diff --git a/docs/index.md b/docs/index.md index bd87695..7bf5397 100644 --- a/docs/index.md +++ b/docs/index.md @@ -13,6 +13,7 @@ tutorial/tutorial-anaerobic-digestion tutorial/demo-mgnify-mgys-1135 +tutorial/demo-mgnify-tomatoes ``` ```{toctree} diff --git a/docs/tutorial/demo-mgnify-mgys-1135.py b/docs/tutorial/demo-mgnify-mgys-1135.py index f65909c..ac7a0c6 100644 --- a/docs/tutorial/demo-mgnify-mgys-1135.py +++ b/docs/tutorial/demo-mgnify-mgys-1135.py @@ -5,7 +5,7 @@ # extension: .py # format_name: percent # format_version: '1.3' -# jupytext_version: 1.17.2 +# jupytext_version: 1.18.1 # kernelspec: # display_name: .venv # language: python diff --git a/docs/tutorial/demo-mgnify-tomatoes.ipynb b/docs/tutorial/demo-mgnify-tomatoes.ipynb new file mode 100644 index 0000000..6949cd6 --- /dev/null +++ b/docs/tutorial/demo-mgnify-tomatoes.ipynb @@ -0,0 +1,7442 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "ad5a71b5", + "metadata": {}, + "source": [ + "# `ABaCo` demo: No biological labels w/ Mgnify tomatoes" + ] + }, + { + "cell_type": "markdown", + "id": "bcc12d4b", + "metadata": {}, + "source": [ + "In this tutorial we demonstrate how ABaCo can be used: \n", + "- without known biological group labels (e.g., only correct batch effect)\n", + "- and for integrating samples from multiple studies.\n", + "\n", + "Specifically, we integrate tomato rhizosphere metagenomics analyses available on MGniFy. \n", + "- MGYS00006231, Illumina HiSeq 4000\n", + "- MGYS00006204, Illumina HiSeq 2500\n", + "- MGYS00006205, Illumina HiSeq 2500\n", + "\n", + "**Additionally** we provide an intro to [working with anndata](https://anndata.readthedocs.io/en/stable/tutorials/notebooks/getting-started.html) \n", + "\n", + "**Note** in this tutorial we include a dummy bio column for demo purposes, but it is not necessary to have a bio column to use abaco. \n", + "\n", + "-----\n", + "**Goal:**\n", + "\n", + "With ABaCo the aim is to remove the technical variation between sequencing platforms. \n", + "\n", + "-----\n", + "\n", + "To start we retrieve the data from MGniFy. " + ] + }, + { + "cell_type": "markdown", + "id": "0ac42400", + "metadata": {}, + "source": [ + "## Downloading the data" + ] + }, + { + "cell_type": "code", + "execution_count": 126, + "id": "f94eae83", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "(189, 12)\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
ERZ12633510ERZ12633501ERZ12633502ERZ12633503ERZ12633504ERZ12633505ERZ12633506ERZ12633507ERZ12633517ERZ12633508ERZ12633518ERZ12633509
taxa
sk__Bacteria;k__;p__Planctomycetes;c__Planctomycetia;o__Pirellulales000110000000
sk__Bacteria;k__;p__Actinobacteria;c__Acidimicrobiia;o__Acidimicrobiales;f__Ilumatobacteraceae;g__Ilumatobacter100000000000
sk__Bacteria;k__;p__Proteobacteria;c__Deltaproteobacteria;o__Myxococcales;f__Kofleriaceae000000000010
sk__Bacteria;k__;p__Proteobacteria;c__Betaproteobacteria;o__Burkholderiales;f__Comamonadaceae000000001000
sk__Bacteria;k__;p__Firmicutes;c__Clostridia;o__Clostridiales;f__Lachnospiraceae001010110000
\n", + "
" + ], + "text/plain": [ + " ERZ12633510 ERZ12633501 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Planctomycetes;c__Planctomy... 0 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Acidimicr... 1 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Betaprote... 0 0 \n", + "sk__Bacteria;k__;p__Firmicutes;c__Clostridia;o_... 0 0 \n", + "\n", + " ERZ12633502 ERZ12633503 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Planctomycetes;c__Planctomy... 0 1 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Acidimicr... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Betaprote... 0 0 \n", + "sk__Bacteria;k__;p__Firmicutes;c__Clostridia;o_... 1 0 \n", + "\n", + " ERZ12633504 ERZ12633505 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Planctomycetes;c__Planctomy... 1 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Acidimicr... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Betaprote... 0 0 \n", + "sk__Bacteria;k__;p__Firmicutes;c__Clostridia;o_... 1 0 \n", + "\n", + " ERZ12633506 ERZ12633507 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Planctomycetes;c__Planctomy... 0 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Acidimicr... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Betaprote... 0 0 \n", + "sk__Bacteria;k__;p__Firmicutes;c__Clostridia;o_... 1 1 \n", + "\n", + " ERZ12633517 ERZ12633508 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Planctomycetes;c__Planctomy... 0 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Acidimicr... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Betaprote... 1 0 \n", + "sk__Bacteria;k__;p__Firmicutes;c__Clostridia;o_... 0 0 \n", + "\n", + " ERZ12633518 ERZ12633509 \n", + "taxa \n", + "sk__Bacteria;k__;p__Planctomycetes;c__Planctomy... 0 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Acidimicr... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 1 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Betaprote... 0 0 \n", + "sk__Bacteria;k__;p__Firmicutes;c__Clostridia;o_... 0 0 " + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "(446, 114)\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
ERZ12343720ERZ12343730ERZ12343740ERZ12343750ERZ12343760ERZ12343770ERZ12343780ERZ12343721ERZ12343731ERZ12343741...ERZ12590947ERZ12593067ERZ12590878ERZ12590908ERZ12590968ERZ12590988ERZ12590849ERZ12590879ERZ12590889ERZ12590909
taxa
sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae;o__Opitutales;f__Opitutaceae;g__Opitutus0000001001...0000010001
sk__Bacteria;k__;p__Candidatus_Saccharibacteria;c__;o__;f__;g__Candidatus_Saccharimonas0000010000...0000000011
sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Entomophthoromycetes0100000000...0000000000
sk__Bacteria;k__;p__Proteobacteria;c__Gammaproteobacteria;o__Xanthomonadales;f__Xanthomonadaceae;g__Pseudoxanthomonas0000000000...0100000000
sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia;o__Herpetosiphonales;f__Herpetosiphonaceae0000000000...0000000000
\n", + "

5 rows × 114 columns

\n", + "
" + ], + "text/plain": [ + " ERZ12343720 ERZ12343730 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 0 0 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 0 0 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 1 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 0 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 0 \n", + "\n", + " ERZ12343740 ERZ12343750 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 0 0 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 0 0 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 0 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 0 \n", + "\n", + " ERZ12343760 ERZ12343770 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 0 0 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 0 1 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 0 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 0 \n", + "\n", + " ERZ12343780 ERZ12343721 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 1 0 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 0 0 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 0 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 0 \n", + "\n", + " ERZ12343731 ERZ12343741 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 0 1 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 0 0 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 0 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 0 \n", + "\n", + " ... ERZ12590947 \\\n", + "taxa ... \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... ... 0 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... ... 0 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... ... 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... ... 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... ... 0 \n", + "\n", + " ERZ12593067 ERZ12590878 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 0 0 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 0 0 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 1 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 0 \n", + "\n", + " ERZ12590908 ERZ12590968 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 0 0 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 0 0 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 0 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 0 \n", + "\n", + " ERZ12590988 ERZ12590849 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 1 0 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 0 0 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 0 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 0 \n", + "\n", + " ERZ12590879 ERZ12590889 \\\n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 0 0 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 0 1 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 0 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 0 \n", + "\n", + " ERZ12590909 \n", + "taxa \n", + "sk__Bacteria;k__;p__Verrucomicrobia;c__Opitutae... 1 \n", + "sk__Bacteria;k__;p__Candidatus_Saccharibacteria... 1 \n", + "sk__Eukaryota;k__Fungi;p__Zoopagomycota;c__Ento... 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Gammaprot... 0 \n", + "sk__Bacteria;k__;p__Chloroflexi;c__Chloroflexia... 0 \n", + "\n", + "[5 rows x 114 columns]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "(188, 12)\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
ERZ12633490ERZ12633491ERZ12633492ERZ12633493ERZ12633494ERZ12633495ERZ12633496ERZ12633497ERZ12633488ERZ12633498ERZ12633489ERZ12633499
taxa
sk__Eukaryota;k__Metazoa;p__Nematoda;c__Chromadorea;o__Rhabditida;f__Cephalobidae011011101000
sk__Bacteria;k__;p__Acidobacteria;c__Blastocatellia;o__Blastocatellales;f__Pyrinomonadaceae111111222111
sk__Bacteria;k__;p__Bacteroidetes;c__Flavobacteriia;o__Flavobacteriales;f__Flavobacteriaceae100000001000
sk__Bacteria;k__;p__Proteobacteria;c__Deltaproteobacteria;o__Myxococcales;f__Archangiaceae200000000010
sk__Bacteria;k__;p__Actinobacteria;c__Actinobacteria;o__Propionibacteriales;f__Propionibacteriaceae;g__Microlunatus000100000011
\n", + "
" + ], + "text/plain": [ + " ERZ12633490 ERZ12633491 \\\n", + "taxa \n", + "sk__Eukaryota;k__Metazoa;p__Nematoda;c__Chromad... 0 1 \n", + "sk__Bacteria;k__;p__Acidobacteria;c__Blastocate... 1 1 \n", + "sk__Bacteria;k__;p__Bacteroidetes;c__Flavobacte... 1 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 2 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Actinobac... 0 0 \n", + "\n", + " ERZ12633492 ERZ12633493 \\\n", + "taxa \n", + "sk__Eukaryota;k__Metazoa;p__Nematoda;c__Chromad... 1 0 \n", + "sk__Bacteria;k__;p__Acidobacteria;c__Blastocate... 1 1 \n", + "sk__Bacteria;k__;p__Bacteroidetes;c__Flavobacte... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 0 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Actinobac... 0 1 \n", + "\n", + " ERZ12633494 ERZ12633495 \\\n", + "taxa \n", + "sk__Eukaryota;k__Metazoa;p__Nematoda;c__Chromad... 1 1 \n", + "sk__Bacteria;k__;p__Acidobacteria;c__Blastocate... 1 1 \n", + "sk__Bacteria;k__;p__Bacteroidetes;c__Flavobacte... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 0 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Actinobac... 0 0 \n", + "\n", + " ERZ12633496 ERZ12633497 \\\n", + "taxa \n", + "sk__Eukaryota;k__Metazoa;p__Nematoda;c__Chromad... 1 0 \n", + "sk__Bacteria;k__;p__Acidobacteria;c__Blastocate... 2 2 \n", + "sk__Bacteria;k__;p__Bacteroidetes;c__Flavobacte... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 0 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Actinobac... 0 0 \n", + "\n", + " ERZ12633488 ERZ12633498 \\\n", + "taxa \n", + "sk__Eukaryota;k__Metazoa;p__Nematoda;c__Chromad... 1 0 \n", + "sk__Bacteria;k__;p__Acidobacteria;c__Blastocate... 2 1 \n", + "sk__Bacteria;k__;p__Bacteroidetes;c__Flavobacte... 1 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 0 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Actinobac... 0 0 \n", + "\n", + " ERZ12633489 ERZ12633499 \n", + "taxa \n", + "sk__Eukaryota;k__Metazoa;p__Nematoda;c__Chromad... 0 0 \n", + "sk__Bacteria;k__;p__Acidobacteria;c__Blastocate... 1 1 \n", + "sk__Bacteria;k__;p__Bacteroidetes;c__Flavobacte... 0 0 \n", + "sk__Bacteria;k__;p__Proteobacteria;c__Deltaprot... 1 0 \n", + "sk__Bacteria;k__;p__Actinobacteria;c__Actinobac... 1 1 " + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "import requests \n", + "from io import StringIO\n", + "import pandas as pd\n", + "import numpy as np\n", + "\n", + "# urls manually retrieved from mgnify website\n", + "datasets = {\n", + " \"MGYS00006205\": \"https://www.ebi.ac.uk/metagenomics/api/v1/studies/MGYS00006205/pipelines/5.0/file/ERP140107_taxonomy_abundances_SSU_v5.0.tsv\",\n", + " \"MGYS00006231\": \"https://www.ebi.ac.uk/metagenomics/api/v1/studies/MGYS00006231/pipelines/5.0/file/ERP139927_taxonomy_abundances_SSU_v5.0.tsv\",\n", + " \"MGYS00006204\": \"https://www.ebi.ac.uk/metagenomics/api/v1/studies/MGYS00006204/pipelines/5.0/file/ERP140102_taxonomy_abundances_SSU_v5.0.tsv\",\n", + "}\n", + "# init for dfs\n", + "dfs = {}\n", + "for key, url in datasets.items():\n", + " # download\n", + " response = requests.get(url)\n", + " # treat as file\n", + " file_like = StringIO(response.content.decode('utf-8'))\n", + " # read into pd df\n", + " dfs[key] = pd.read_csv(file_like, sep='\\t')\n", + " # rename sampleid col to taxa\n", + " dfs[key] = dfs[key].rename(columns={\"#SampleID\": \"taxa\"})\n", + " # set index to taxa\n", + " dfs[key] = dfs[key].set_index('taxa')\n", + " # drop rows with all zeros\n", + " dfs[key] = dfs[key][dfs[key].sum(axis=1)>0]\n", + " # sanity check\n", + " print(dfs[key].shape)\n", + " display(dfs[key].sample(5))" + ] + }, + { + "cell_type": "markdown", + "id": "6a035933", + "metadata": {}, + "source": [ + "Based on the sampled outputs printed above, w can see that the data: \n", + "- has taxa as rows\n", + "- the samples as columns \n", + "- and the values are the number of taxanomic assignments for the given taxa x sample\n", + "\n", + "To join them we will use a helper function `abaco.utils.df_joiner()`. " + ] + }, + { + "cell_type": "code", + "execution_count": 153, + "id": "c6d2108b", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\n", + "Index: 531 entries, sk__Archaea to sk__Eukaryota;k__Viridiplantae;p__Streptophyta;c__Magnoliopsida;o__Solanales;f__Solanaceae\n", + "Columns: 138 entries, ERZ12633510 to ERZ12633499\n", + "dtypes: float64(138)\n", + "memory usage: 576.6+ KB\n" + ] + } + ], + "source": [ + "from abaco.utils import df_joiner\n", + "\n", + "joined = df_joiner(\n", + " df_dict=dfs, \n", + " on=\"taxa\",\n", + " how=\"outer\",\n", + ")\n", + "\n", + "joined.info()" + ] + }, + { + "cell_type": "markdown", + "id": "9511c611", + "metadata": {}, + "source": [ + "## A brief intro to `anndata`\n", + "\n", + "![img](https://raw.githubusercontent.com/scverse/anndata/main/docs/_static/img/anndata_schema.svg)\n", + "\n", + "For this example we will use anndata to store multiple layers of the data (e.g., raw counts, normalized, etc.) and annotate the samples and taxa with metadata. Below we prepare the annotations (i.e., obs, var). " + ] + }, + { + "cell_type": "code", + "execution_count": 154, + "id": "b6303749", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "(138, 2)\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
taxasourcebio
ERZ12343766MGYS00006231Group1
ERZ12590853MGYS00006231Group2
ERZ12343727MGYS00006231Group2
ERZ12343729MGYS00006231Group1
ERZ12590922MGYS00006231Group2
\n", + "
" + ], + "text/plain": [ + "taxa source bio\n", + "ERZ12343766 MGYS00006231 Group1\n", + "ERZ12590853 MGYS00006231 Group2\n", + "ERZ12343727 MGYS00006231 Group2\n", + "ERZ12343729 MGYS00006231 Group1\n", + "ERZ12590922 MGYS00006231 Group2" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "(531, 8)\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
superkingdomkingdomphylumclassorderfamilygenusspecies
175sk__Bacteriak__p__Candidatus_Omnitrophicac__o__f__g__s__Omnitrophica_WOR_2_bacterium_RBG_13_44_8
478sk__Eukaryotak__p__Haptistac__Centroplasthelidao__Pterocystidaf__g__ChoanocystisNone
219sk__Bacteriak__p__Fibrobacteresc__ChitinivibrioniaNoneNoneNoneNone
165sk__Bacteriak__p__Candidatus_DoudnabacteriaNoneNoneNoneNoneNone
369sk__Bacteriak__p__Proteobacteriac__Deltaproteobacteriao__Myxococcalesf__Kofleriaceaeg__HaliangiumNone
\n", + "
" + ], + "text/plain": [ + " superkingdom kingdom phylum \\\n", + "175 sk__Bacteria k__ p__Candidatus_Omnitrophica \n", + "478 sk__Eukaryota k__ p__Haptista \n", + "219 sk__Bacteria k__ p__Fibrobacteres \n", + "165 sk__Bacteria k__ p__Candidatus_Doudnabacteria \n", + "369 sk__Bacteria k__ p__Proteobacteria \n", + "\n", + " class order family \\\n", + "175 c__ o__ f__ \n", + "478 c__Centroplasthelida o__Pterocystida f__ \n", + "219 c__Chitinivibrionia None None \n", + "165 None None None \n", + "369 c__Deltaproteobacteria o__Myxococcales f__Kofleriaceae \n", + "\n", + " genus species \n", + "175 g__ s__Omnitrophica_WOR_2_bacterium_RBG_13_44_8 \n", + "478 g__Choanocystis None \n", + "219 None None \n", + "165 None None \n", + "369 g__Haliangium None " + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "import anndata as ad\n", + "import scanpy as sc\n", + "\n", + "# preparing obs metadata\n", + "dfs_T = {}\n", + "# for each study\n", + "for key in dfs:\n", + " # transpose\n", + " dfs_T[key] = dfs[key].T\n", + " # add study source column to be batch effect\n", + " dfs_T[key]['source'] = key\n", + " # dummy bio column, rando group assign, for checking and demo \n", + " dfs_T[key]['bio'] = np.random.choice([\"Group1\", \"Group2\"], size = len(dfs_T[key]))\n", + " # drop allother cols\n", + " dfs_T[key] = dfs_T[key][['source', 'bio']]\n", + "# now concat all study metadata \n", + "obs = pd.concat(dfs_T.values())\n", + "print(obs.shape)\n", + "display(obs.sample(5))\n", + "\n", + "# preparing var metadata (taxa levels)\n", + "# get taxa index as df and split on ';'\n", + "var = joined.reset_index()[['taxa']]\n", + "var = var['taxa'].str.split(';', expand=True)\n", + "# rename cols \n", + "var.columns = [\n", + " \"superkingdom\", \"kingdom\",\"phylum\", \"class\", \n", + " \"order\", \"family\", \"genus\", \"species\"\n", + "]\n", + "print(var.shape)\n", + "display(var.sample(5))" + ] + }, + { + "cell_type": "markdown", + "id": "62e44f01", + "metadata": {}, + "source": [ + "We can initiate the AnnData objct with the values (taxanomic assignment counts) and the metadata we prepped above. " + ] + }, + { + "cell_type": "code", + "execution_count": 155, + "id": "d98ecaf7", + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/Users/anglup/.pyenv/versions/3.11.0/lib/python3.11/functools.py:909: ImplicitModificationWarning:\n", + "\n", + "Transforming to str index.\n", + "\n" + ] + }, + { + "data": { + "text/plain": [ + "AnnData object with n_obs × n_vars = 138 × 531\n", + " obs: 'source', 'bio'\n", + " var: 'superkingdom', 'kingdom', 'phylum', 'class', 'order', 'family', 'genus', 'species'" + ] + }, + "execution_count": 155, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "# init anndata object\n", + "full = ad.AnnData(\n", + " X=joined.T.values,\n", + " obs=obs,\n", + " var=var, \n", + ")\n", + "# check it out\n", + "full" + ] + }, + { + "cell_type": "markdown", + "id": "7d17b0bd", + "metadata": {}, + "source": [ + "With anndata we can also store additional layers seamlessly with the same annoations. Below we will create a layer with the NaNs filled with 0s" + ] + }, + { + "cell_type": "code", + "execution_count": 156, + "id": "d272c7b1", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "AnnData object with n_obs × n_vars = 138 × 531\n", + " obs: 'source', 'bio'\n", + " var: 'superkingdom', 'kingdom', 'phylum', 'class', 'order', 'family', 'genus', 'species'\n", + " layers: 'nona'\n", + "original:\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
0123456789...521522523524525526527528529530
ERZ126335103.0NaNNaNNaN0.00.00.0NaN1.02.0...NaNNaNNaNNaN1.0NaNNaN0.0NaNNaN
ERZ126335010.0NaNNaNNaN0.00.00.0NaN0.02.0...NaNNaNNaNNaN0.0NaNNaN0.0NaNNaN
ERZ126335022.0NaNNaNNaN1.00.01.0NaN0.04.0...NaNNaNNaNNaN0.0NaNNaN0.0NaNNaN
ERZ126335033.0NaNNaNNaN0.00.00.0NaN0.01.0...NaNNaNNaNNaN0.0NaNNaN0.0NaNNaN
ERZ126335040.0NaNNaNNaN0.00.00.0NaN0.02.0...NaNNaNNaNNaN0.0NaNNaN0.0NaNNaN
\n", + "

5 rows × 531 columns

\n", + "
" + ], + "text/plain": [ + " 0 1 2 3 4 5 6 7 8 9 ... 521 522 523 \\\n", + "ERZ12633510 3.0 NaN NaN NaN 0.0 0.0 0.0 NaN 1.0 2.0 ... NaN NaN NaN \n", + "ERZ12633501 0.0 NaN NaN NaN 0.0 0.0 0.0 NaN 0.0 2.0 ... NaN NaN NaN \n", + "ERZ12633502 2.0 NaN NaN NaN 1.0 0.0 1.0 NaN 0.0 4.0 ... NaN NaN NaN \n", + "ERZ12633503 3.0 NaN NaN NaN 0.0 0.0 0.0 NaN 0.0 1.0 ... NaN NaN NaN \n", + "ERZ12633504 0.0 NaN NaN NaN 0.0 0.0 0.0 NaN 0.0 2.0 ... NaN NaN NaN \n", + "\n", + " 524 525 526 527 528 529 530 \n", + "ERZ12633510 NaN 1.0 NaN NaN 0.0 NaN NaN \n", + "ERZ12633501 NaN 0.0 NaN NaN 0.0 NaN NaN \n", + "ERZ12633502 NaN 0.0 NaN NaN 0.0 NaN NaN \n", + "ERZ12633503 NaN 0.0 NaN NaN 0.0 NaN NaN \n", + "ERZ12633504 NaN 0.0 NaN NaN 0.0 NaN NaN \n", + "\n", + "[5 rows x 531 columns]" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stdout", + "output_type": "stream", + "text": [ + "no na:\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
0123456789...521522523524525526527528529530
ERZ126335103.00.00.00.00.00.00.00.01.02.0...0.00.00.00.01.00.00.00.00.00.0
ERZ126335010.00.00.00.00.00.00.00.00.02.0...0.00.00.00.00.00.00.00.00.00.0
ERZ126335022.00.00.00.01.00.01.00.00.04.0...0.00.00.00.00.00.00.00.00.00.0
ERZ126335033.00.00.00.00.00.00.00.00.01.0...0.00.00.00.00.00.00.00.00.00.0
ERZ126335040.00.00.00.00.00.00.00.00.02.0...0.00.00.00.00.00.00.00.00.00.0
\n", + "

5 rows × 531 columns

\n", + "
" + ], + "text/plain": [ + " 0 1 2 3 4 5 6 7 8 9 ... 521 522 \\\n", + "ERZ12633510 3.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 2.0 ... 0.0 0.0 \n", + "ERZ12633501 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 2.0 ... 0.0 0.0 \n", + "ERZ12633502 2.0 0.0 0.0 0.0 1.0 0.0 1.0 0.0 0.0 4.0 ... 0.0 0.0 \n", + "ERZ12633503 3.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 ... 0.0 0.0 \n", + "ERZ12633504 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 2.0 ... 0.0 0.0 \n", + "\n", + " 523 524 525 526 527 528 529 530 \n", + "ERZ12633510 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 \n", + "ERZ12633501 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 \n", + "ERZ12633502 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 \n", + "ERZ12633503 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 \n", + "ERZ12633504 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 \n", + "\n", + "[5 rows x 531 columns]" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "# layer with no na\n", + "full.layers['nona'] = np.nan_to_num(full.X, nan=0)\n", + "print(full)\n", + "\n", + "# checking it out, also demoing to_df method\n", + "print(\"original:\")\n", + "display(full.to_df().head(5))\n", + "print(\"no na:\")\n", + "display(full.to_df(layer='nona').head(5))" + ] + }, + { + "cell_type": "markdown", + "id": "4f31ba7c", + "metadata": {}, + "source": [ + "we can also add normalized and clr layers. For more details on the reasoning of these layers visit our other tutorials." + ] + }, + { + "cell_type": "code", + "execution_count": 157, + "id": "2300de51", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "AnnData object with n_obs × n_vars = 138 × 531\n", + " obs: 'source', 'bio'\n", + " var: 'superkingdom', 'kingdom', 'phylum', 'class', 'order', 'family', 'genus', 'species'\n", + " layers: 'nona', 'norm', 'clr'\n" + ] + } + ], + "source": [ + "from skbio.stats.composition import clr \n", + "\n", + "full.layers['norm'] = full.layers['nona']/full.layers['nona'].sum(axis=1).reshape(-1,1)\n", + "\n", + "# CLR transform normalized data, replacing zeros with smallll val to avoid log(0)\n", + "full.layers['clr'] = clr(np.where(full.layers['norm'] > 0, full.layers['norm'], 1e-10))\n", + "\n", + "print(full)" + ] + }, + { + "cell_type": "markdown", + "id": "205062a7", + "metadata": {}, + "source": [ + "With anndata we can also filter the data using the annotations/metadata. Below we will keep the variables (taxa) with genus assignment. Additionally we filter out some taxa if they are missing in more than 30% of the samples. " + ] + }, + { + "cell_type": "code", + "execution_count": 158, + "id": "a6d0c0f7", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "View of AnnData object with n_obs × n_vars = 138 × 220\n", + " obs: 'source', 'bio'\n", + " var: 'superkingdom', 'kingdom', 'phylum', 'class', 'order', 'family', 'genus', 'species'\n", + " layers: 'nona', 'norm', 'clr'\n", + "View of AnnData object with n_obs × n_vars = 138 × 176\n", + " obs: 'source', 'bio'\n", + " var: 'superkingdom', 'kingdom', 'phylum', 'class', 'order', 'family', 'genus', 'species'\n", + " layers: 'nona', 'norm', 'clr'\n" + ] + } + ], + "source": [ + "# must have genus level\n", + "adata = full[:, ~full.var['genus'].isna()]\n", + "print(adata)\n", + "\n", + "# filtering if taxa is missing (NaN) in more than 30% of samples\n", + "adata = adata[:, np.isnan(adata.X).sum(axis=0)/adata.X.shape[0] <= 0.3]\n", + "print(adata)" + ] + }, + { + "cell_type": "markdown", + "id": "4ba030e0", + "metadata": {}, + "source": [ + "We are down to 176 taxa. \n", + "\n", + "However, it can be trickier to aggregate with anndata -- but we can use the [scanpy toolkit](https://scanpy.readthedocs.io/en/stable/index.html) to help. \n", + "\n", + "Specifically we will use [scanpy.get.aggregate()](https://scanpy.readthedocs.io/en/stable/generated/scanpy.get.aggregate.html) to help us check that the dummy biological groups we created are not different. " + ] + }, + { + "cell_type": "code", + "execution_count": 161, + "id": "f47a2606", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "AnnData object with n_obs × n_vars = 2 × 176\n", + " obs: 'bio'\n", + " var: 'superkingdom', 'kingdom', 'phylum', 'class', 'order', 'family', 'genus', 'species'\n", + " layers: 'mean'\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
11122021232631323337...466468470476477478488489497527
Group1-0.656399-2.371363-2.117223-2.125450-2.122299-1.382745-0.883526-2.371363-1.3824691.598991...1.300304-1.892162-2.371363-2.124440-2.371363-2.371363-2.122834-2.126390-2.122834-2.371363
Group20.064966-2.202220-1.344149-2.475478-2.190108-2.189974-0.786739-1.912433-0.4818740.379781...0.599701-2.475478-2.197352-2.475478-2.194094-2.202220-1.905638-2.189702-2.475478-2.189974
\n", + "

2 rows × 176 columns

\n", + "
" + ], + "text/plain": [ + " 11 12 20 21 23 26 31 \\\n", + "Group1 -0.656399 -2.371363 -2.117223 -2.125450 -2.122299 -1.382745 -0.883526 \n", + "Group2 0.064966 -2.202220 -1.344149 -2.475478 -2.190108 -2.189974 -0.786739 \n", + "\n", + " 32 33 37 ... 466 468 470 \\\n", + "Group1 -2.371363 -1.382469 1.598991 ... 1.300304 -1.892162 -2.371363 \n", + "Group2 -1.912433 -0.481874 0.379781 ... 0.599701 -2.475478 -2.197352 \n", + "\n", + " 476 477 478 488 489 497 527 \n", + "Group1 -2.124440 -2.371363 -2.371363 -2.122834 -2.126390 -2.122834 -2.371363 \n", + "Group2 -2.475478 -2.194094 -2.202220 -1.905638 -2.189702 -2.475478 -2.189974 \n", + "\n", + "[2 rows x 176 columns]" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "# agg by fake bio group \n", + "agg_mean = sc.get.aggregate(\n", + " # the anndata object\n", + " adata=adata,\n", + " # what layer to use\n", + " layer='clr',\n", + " # group by col\n", + " by=\"bio\",\n", + " # axis that the group lives\n", + " axis=\"obs\",\n", + " # how to agg\n", + " func='mean',\n", + ")\n", + "# check out the returned AnnData objct\n", + "print(agg_mean)\n", + "# we can see taht the default layer name if not provided is the func name \n", + "display(agg_mean.to_df('mean').head(5))" + ] + }, + { + "cell_type": "markdown", + "id": "e1f33e18", + "metadata": {}, + "source": [ + "Now lets quickly check that group 1 and 2 are not different visually and quick stats." + ] + }, + { + "cell_type": "code", + "execution_count": 162, + "id": "80fcd19b", + "metadata": {}, + "outputs": [ + { + "data": { + "text/html": [ + "
" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/plain": [ + "WilcoxonResult(statistic=np.float64(7108.0), pvalue=np.float64(0.3150957187537212))" + ] + }, + "execution_count": 162, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "import plotly.graph_objects as go\n", + "from scipy import stats\n", + "\n", + "# transpose for easier plotting\n", + "bio_sum = agg_mean.to_df('mean').T\n", + "\n", + "# init fig\n", + "fig = go.Figure()\n", + "# add traces\n", + "fig.add_trace(go.Histogram(x=bio_sum['Group1'], name='Group1'))\n", + "fig.add_trace(go.Histogram(x=bio_sum['Group2'], name='Group2'))\n", + "\n", + "# formatting\n", + "fig.update_layout(barmode='overlay')\n", + "fig.update_traces(opacity=0.75)\n", + "fig.show()\n", + "\n", + "# check no diff in stats \n", + "stats.wilcoxon(bio_sum['Group1'], bio_sum['Group2'])" + ] + }, + { + "cell_type": "markdown", + "id": "66663906", + "metadata": {}, + "source": [ + "## Prepare data for ABaCo\n", + "\n", + "Great. Now lets proceed with adata where we can easily extract a df to meet the data format required for abaco." + ] + }, + { + "cell_type": "code", + "execution_count": 163, + "id": "1d77b8ab", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\n", + "RangeIndex: 138 entries, 0 to 137\n", + "Columns: 179 entries, index to bio\n", + "dtypes: category(2), float64(176), object(1)\n", + "memory usage: 191.5+ KB\n" + ] + } + ], + "source": [ + "batch_col = 'source'\n", + "bio_col = 'bio'\n", + "id_col = 'index'\n", + "\n", + "# the raw counts as df\n", + "df_taxa = adata.to_df(\"nona\")\n", + "\n", + "# appending on the categorical data of interest\n", + "df_all = pd.concat([df_taxa, adata.obs[[batch_col, bio_col]]], axis=1).reset_index()\n", + "\n", + "df_all['bio'] = df_all['bio'].astype('category')\n", + "df_all['source'] = df_all['source'].astype('category')\n", + "\n", + "df_all.info()" + ] + }, + { + "cell_type": "markdown", + "id": "79dfcc4f", + "metadata": {}, + "source": [ + "**`pd.DataFrame` Requirements for ABaCo:**\n", + "\n", + "The dataset contains the following making it compatible with the ABaCo framework:\n", + "\n", + "| id_col | batch_col | bio_col | count1 | count2 | ... |\n", + "|--------|------------|----------|--------|--------|-----|\n", + "| A | 24/07/2025 | RA | # | # | ... |\n", + "| B | 15/06/2024 | RD | # | # | ... |\n", + "| C | 24/07/2025 | RL | # | # | ... |\n", + "\n", + "- The data has categorical columns: \n", + " 1. unique ids to identify the observations/samples e.g. sample id col\n", + " 2. ids for the batch/factor groupings to be corrected by abaco. e.g. our phony bio data \n", + " 3. biological/experimental factor variation for abaco to retain when correcting batch effect e.g., study id \n", + "\n", + "- And taxa counts to be trained on. \n", + "\n", + "We can use `abaco.plots.plotPCoA()` to visualize any batch and biological effects based on the given categories. " + ] + }, + { + "cell_type": "code", + "execution_count": 164, + "id": "128ba953", + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/Users/anglup/GitHub/abaco/.venv/lib/python3.11/site-packages/skbio/stats/ordination/_principal_coordinate_analysis.py:157: RuntimeWarning:\n", + "\n", + "EIGH: since no value for dimensions is specified, PCoA for all dimensions will be computed, which may result in long computation time if the original distance matrix is large.\n", + "\n" + ] + }, + { + "data": { + "text/html": [ + " \n", + " \n", + " " + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "from abaco.plots import plotPCoA\n", + "import plotly.io as pio\n", + "pio.renderers.default = \"notebook\"\n", + "\n", + "plotPCoA(\n", + " data=df_all, \n", + " sample_label=id_col, \n", + " batch_label=batch_col, \n", + " experiment_label=bio_col,\n", + ")" + ] + }, + { + "cell_type": "markdown", + "id": "ddbd9118", + "metadata": {}, + "source": [ + "- Batch effect (colours): \n", + " - Different sequencing platforms which could result in a technical source of variation captured by the clustering of studies 6204 & 6205 (Illumina HiSeq 2500) vs. study 6231 (Illumina HiSeq 4000) along PCo1\n", + "\n", + "- Biological effect (shapes): \n", + " - The groups were randomly assigned so we expect no clustering by shape as supported by the pcoa above. This behaviour should not change with abaco reconstruction.\n", + "-----" + ] + }, + { + "cell_type": "markdown", + "id": "bbae141c", + "metadata": {}, + "source": [ + "## The goal \n", + "\n", + "Here the aim of **ABaCo** is to: \n", + "1) correct the batch effect (e.g., the points should no longer cluster by colour in the PCoA) while\n", + "2) maintaining biological variance (or lack thereof).\n", + "\n", + "Ideally, after using AbaCo to transform the data, the resulting PCoA coloured by batches will look like a colourful mixture of points.\n", + "\n", + "-----\n", + "\n", + "## Using `ABaCo`\n", + "\n", + "### Setting up ABaCo\n", + "\n", + "We instantiate the `abaco.metaABaCo()` class and pass the required parameters shown in the cell below. \n", + "\n", + "Usually, setup of the parameters is required, which are explained in brief in the documentation e.g. `help(metaABaCo)`\n", + "\n" + ] + }, + { + "cell_type": "code", + "execution_count": 264, + "id": "ac1c6cc7", + "metadata": {}, + "outputs": [], + "source": [ + "from abaco.ABaCo import metaABaCo\n", + "import torch\n", + "\n", + "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n", + "\n", + "# Instaantiate the ABaCo model\n", + "model = metaABaCo(\n", + " data=df_all, # Pre-processed dataframe\n", + " n_bios=df_all[bio_col].nunique(), # Number of biological groups in the data\n", + " bio_label=bio_col, # Column where biological groups are labeled in the dataframe\n", + " n_batches=df_all[batch_col].nunique(), # Number of batch groups in the data\n", + " batch_label=batch_col, # Column where batch groups are labeled in the dataframe\n", + " n_features=df_taxa.shape[1], # Number of features (taxonomic groups)\n", + " prior=\"MoG\", # Prior distribution \n", + " device=device, # Device\n", + " d_z=16, # default dim of latent space\n", + " epochs = [3000, 1000, 3000], # num epochs for each training phase\n", + " disc_net=[256, 128, 64], # stronger discriminator\n", + " disc_act_fun=torch.nn.LeakyReLU(0.1),\n", + ")" + ] + }, + { + "cell_type": "code", + "execution_count": 265, + "id": "2f20940c", + "metadata": { + "tags": [ + "hide-output" + ] + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Help on class metaABaCo in module abaco.ABaCo:\n", + "\n", + "class metaABaCo(torch.nn.modules.module.Module)\n", + " | metaABaCo(data, n_bios, bio_label, n_batches, batch_label, n_features, device, prior='MoG', pdist='ZINB', d_z=16, epochs=[1000, 2000, 2000], encoder_net=[512, 256, 128], decoder_net=[128, 256, 512], vae_act_fun=ReLU(), disc_net=[128, 64], disc_act_fun=ReLU())\n", + " | \n", + " | Method resolution order:\n", + " | metaABaCo\n", + " | torch.nn.modules.module.Module\n", + " | builtins.object\n", + " | \n", + " | Methods defined here:\n", + " | \n", + " | __init__(self, data, n_bios, bio_label, n_batches, batch_label, n_features, device, prior='MoG', pdist='ZINB', d_z=16, epochs=[1000, 2000, 2000], encoder_net=[512, 256, 128], decoder_net=[128, 256, 512], vae_act_fun=ReLU(), disc_net=[128, 64], disc_act_fun=ReLU())\n", + " | Function to create the metaABaCo model.\n", + " | \n", + " | Parameters\n", + " | ----------\n", + " | data: pd.DataFrame\n", + " | Pre-processed DataFrame to correct. Only feature columns to correct should be of numerical data type.\n", + " | n_bios: int\n", + " | Number of labels or (potential) clusters based on biological variance. For example, if 2 experimental\n", + " | conditions (e.g., control and treatment) then n_bios = 2.\n", + " | bio_label: char\n", + " | Column label where biological labels are contained in data.\n", + " | n_batches: int\n", + " | Number of batches in the dataset. For example, if samples were sequenced in\n", + " | 5 batches (e.g., 5 different dates) then batches = 5.\n", + " | batch_label: char\n", + " | Column label where batch labels are contained in data.\n", + " | n_features: int\n", + " | Number of features in the input data, columns. For example, if the input is a gene expression matrix with 1000 genes,\n", + " | then input_size = 1000.\n", + " | device: torch.device\n", + " | Device to run the model on, e.g., \"cuda\" or \"cpu\".\n", + " | prior: str\n", + " | Prior distribution used. Baseline is \"VMM\" (VampPrior Mixture Model). Options are \"VMM\" and\n", + " | \"MoG\" (Mixture of Gaussians).\n", + " | pdist: str\n", + " | Output distribution used. Baseline is \"ZINB\" (Zero-inflated Negative Binomial).\n", + " | d_z: int\n", + " | Dimensionality of the latent space. For example, if d_z = 16, then the latent space will have 16 dimensions.\n", + " | epochs: list\n", + " | Number of epochs for first, second and third phase of ABaCo. Default is [1000, 2000, 2000]\n", + " | encoder_net: list\n", + " | List of integers defining the architecture of the encoder. Each integer is a layer size.\n", + " | For example, [1024, 512, 256] means the encoder will have three layers with 1024, 512, and 256 neurons respectively.\n", + " | decoder_net: list\n", + " | List of integers defining the architecture of the decoder. Each integer is a layer size.\n", + " | For example, [256, 512, 1024] means the decoder will have three layers with 256, 512, and 1024 neurons respectively.\n", + " | vae_act_func: nn.Module\n", + " | Activation function for the VAE encoder and decoder. Default is nn.ReLU().\n", + " | disc_net: list\n", + " | List of integers defining the architecture of the discriminator. Each integer is a layer size.\n", + " | For example, [256, 128, 64] means the discriminator will have three layers with 256, 128, and 64 neurons respectively.\n", + " | disc_act_fun: nn.Module\n", + " | Activation function for the discriminator. Default is nn.ReLU().\n", + " | \n", + " | batch_correct(self, train_loader, vae_optimizer, disc_optimizer, adv_optimizer, w_disc=1.0, w_adv=1.0, w_elbo_nll=1.0, w_elbo_kl=1.0, w_bio_penalty=1.0, w_cluster_penalty=1.0)\n", + " | Train the conditional VAE model for batch correction. This is trained after VAE prior parameters are defined,\n", + " | \n", + " | batch_mask(self, train_loader, decoder_optimizer, smooth_annealing=True, cycle_reg=None, w_elbo_nll=1.0, w_cycle=0.001)\n", + " | Pre-trained VAE will now have frozen encoder and batch labels masked at the encoder.\n", + " | \n", + " | correct(self, seed=None, mask=True)\n", + " | \n", + " | fit(self, smooth_annealing=True, cycle_reg=None, seed=None, w_elbo_nll=1.0, w_elbo_kl=1.0, w_bio_penalty=1.0, w_cluster_penalty=1.0, w_cycle=0.001, w_disc=1.0, w_adv=1.0, phase_1_vae_lr=0.001, phase_2_vae_lr=0.001, phase_3_vae_lr=1e-06, disc_lr=0.001, adv_lr=0.001)\n", + " | \n", + " | plot_pca_posterior(self, figsize=(14, 6), palette='tab10')\n", + " | Get the plot of the first 2 principal components of the posterior distribution.\n", + " | \n", + " | train_vae(self, train_loader, optimizer, w_elbo_nll=1.0, w_elbo_kl=1.0, w_bio_penalty=1.0, w_cluster_penalty=1.0)\n", + " | Train the conditional VAE model. If clustering prior is used, penalization term is applied to increase sparsity of the clusters.\n", + " | \n", + " | Parameters:\n", + " | vae: [VAE]\n", + " | Variational Autoencoder model\n", + " | train_loader: [torch.utils.data.DataLoader]\n", + " | DataLoader for the training data\n", + " | optimizer: [torch.optim.Optimizer]\n", + " | Optimizer for training\n", + " | epochs: [int]\n", + " | Number of training epochs\n", + " | device: [str]\n", + " | Device to use for computations\n", + " | \n", + " | ----------------------------------------------------------------------\n", + " | Data and other attributes defined here:\n", + " | \n", + " | __annotations__ = {}\n", + " | \n", + " | ----------------------------------------------------------------------\n", + " | Methods inherited from torch.nn.modules.module.Module:\n", + " | \n", + " | __call__ = _wrapped_call_impl(self, *args, **kwargs)\n", + " | \n", + " | __delattr__(self, name) -> None\n", + " | Implement delattr(self, name).\n", + " | \n", + " | __dir__(self)\n", + " | Default dir() implementation.\n", + " | \n", + " | __getattr__(self, name: str) -> Union[torch.Tensor, ForwardRef('Module')]\n", + " | # It is crucial that the return type is not annotated as `Any`, otherwise type checking\n", + " | # on `torch.nn.Module` and all its subclasses is largely disabled as a result. See:\n", + " | # https://github.com/pytorch/pytorch/pull/115074\n", + " | \n", + " | __getstate__(self)\n", + " | Helper for pickle.\n", + " | \n", + " | __repr__(self) -> str\n", + " | Return repr(self).\n", + " | \n", + " | __setattr__(self, name: str, value: Union[torch.Tensor, ForwardRef('Module')]) -> None\n", + " | Implement setattr(self, name, value).\n", + " | \n", + " | __setstate__(self, state)\n", + " | \n", + " | add_module(self, name: str, module: Optional[ForwardRef('Module')]) -> None\n", + " | Add a child module to the current module.\n", + " | \n", + " | The module can be accessed as an attribute using the given name.\n", + " | \n", + " | Args:\n", + " | name (str): name of the child module. The child module can be\n", + " | accessed from this module using the given name\n", + " | module (Module): child module to be added to the module.\n", + " | \n", + " | apply(self, fn: Callable[[ForwardRef('Module')], NoneType]) -> Self\n", + " | Apply ``fn`` recursively to every submodule (as returned by ``.children()``) as well as self.\n", + " | \n", + " | Typical use includes initializing the parameters of a model\n", + " | (see also :ref:`nn-init-doc`).\n", + " | \n", + " | Args:\n", + " | fn (:class:`Module` -> None): function to be applied to each submodule\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> @torch.no_grad()\n", + " | >>> def init_weights(m):\n", + " | >>> print(m)\n", + " | >>> if type(m) == nn.Linear:\n", + " | >>> m.weight.fill_(1.0)\n", + " | >>> print(m.weight)\n", + " | >>> net = nn.Sequential(nn.Linear(2, 2), nn.Linear(2, 2))\n", + " | >>> net.apply(init_weights)\n", + " | Linear(in_features=2, out_features=2, bias=True)\n", + " | Parameter containing:\n", + " | tensor([[1., 1.],\n", + " | [1., 1.]], requires_grad=True)\n", + " | Linear(in_features=2, out_features=2, bias=True)\n", + " | Parameter containing:\n", + " | tensor([[1., 1.],\n", + " | [1., 1.]], requires_grad=True)\n", + " | Sequential(\n", + " | (0): Linear(in_features=2, out_features=2, bias=True)\n", + " | (1): Linear(in_features=2, out_features=2, bias=True)\n", + " | )\n", + " | \n", + " | bfloat16(self) -> Self\n", + " | Casts all floating point parameters and buffers to ``bfloat16`` datatype.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | buffers(self, recurse: bool = True) -> collections.abc.Iterator[torch.Tensor]\n", + " | Return an iterator over module buffers.\n", + " | \n", + " | Args:\n", + " | recurse (bool): if True, then yields buffers of this module\n", + " | and all submodules. Otherwise, yields only buffers that\n", + " | are direct members of this module.\n", + " | \n", + " | Yields:\n", + " | torch.Tensor: module buffer\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> # xdoctest: +SKIP(\"undefined vars\")\n", + " | >>> for buf in model.buffers():\n", + " | >>> print(type(buf), buf.size())\n", + " | (20L,)\n", + " | (20L, 1L, 5L, 5L)\n", + " | \n", + " | children(self) -> collections.abc.Iterator['Module']\n", + " | Return an iterator over immediate children modules.\n", + " | \n", + " | Yields:\n", + " | Module: a child module\n", + " | \n", + " | compile(self, *args, **kwargs)\n", + " | Compile this Module's forward using :func:`torch.compile`.\n", + " | \n", + " | This Module's `__call__` method is compiled and all arguments are passed as-is\n", + " | to :func:`torch.compile`.\n", + " | \n", + " | See :func:`torch.compile` for details on the arguments for this function.\n", + " | \n", + " | cpu(self) -> Self\n", + " | Move all model parameters and buffers to the CPU.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | cuda(self, device: Union[torch.device, int, NoneType] = None) -> Self\n", + " | Move all model parameters and buffers to the GPU.\n", + " | \n", + " | This also makes associated parameters and buffers different objects. So\n", + " | it should be called before constructing the optimizer if the module will\n", + " | live on GPU while being optimized.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Args:\n", + " | device (int, optional): if specified, all parameters will be\n", + " | copied to that device\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | double(self) -> Self\n", + " | Casts all floating point parameters and buffers to ``double`` datatype.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | eval(self) -> Self\n", + " | Set the module in evaluation mode.\n", + " | \n", + " | This has an effect only on certain modules. See the documentation of\n", + " | particular modules for details of their behaviors in training/evaluation\n", + " | mode, i.e. whether they are affected, e.g. :class:`Dropout`, :class:`BatchNorm`,\n", + " | etc.\n", + " | \n", + " | This is equivalent with :meth:`self.train(False) `.\n", + " | \n", + " | See :ref:`locally-disable-grad-doc` for a comparison between\n", + " | `.eval()` and several similar mechanisms that may be confused with it.\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | extra_repr(self) -> str\n", + " | Return the extra representation of the module.\n", + " | \n", + " | To print customized extra information, you should re-implement\n", + " | this method in your own modules. Both single-line and multi-line\n", + " | strings are acceptable.\n", + " | \n", + " | float(self) -> Self\n", + " | Casts all floating point parameters and buffers to ``float`` datatype.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | forward = _forward_unimplemented(self, *input: Any) -> None\n", + " | Define the computation performed at every call.\n", + " | \n", + " | Should be overridden by all subclasses.\n", + " | \n", + " | .. note::\n", + " | Although the recipe for forward pass needs to be defined within\n", + " | this function, one should call the :class:`Module` instance afterwards\n", + " | instead of this since the former takes care of running the\n", + " | registered hooks while the latter silently ignores them.\n", + " | \n", + " | get_buffer(self, target: str) -> 'Tensor'\n", + " | Return the buffer given by ``target`` if it exists, otherwise throw an error.\n", + " | \n", + " | See the docstring for ``get_submodule`` for a more detailed\n", + " | explanation of this method's functionality as well as how to\n", + " | correctly specify ``target``.\n", + " | \n", + " | Args:\n", + " | target: The fully-qualified string name of the buffer\n", + " | to look for. (See ``get_submodule`` for how to specify a\n", + " | fully-qualified string.)\n", + " | \n", + " | Returns:\n", + " | torch.Tensor: The buffer referenced by ``target``\n", + " | \n", + " | Raises:\n", + " | AttributeError: If the target string references an invalid\n", + " | path or resolves to something that is not a\n", + " | buffer\n", + " | \n", + " | get_extra_state(self) -> Any\n", + " | Return any extra state to include in the module's state_dict.\n", + " | \n", + " | Implement this and a corresponding :func:`set_extra_state` for your module\n", + " | if you need to store extra state. This function is called when building the\n", + " | module's `state_dict()`.\n", + " | \n", + " | Note that extra state should be picklable to ensure working serialization\n", + " | of the state_dict. We only provide backwards compatibility guarantees\n", + " | for serializing Tensors; other objects may break backwards compatibility if\n", + " | their serialized pickled form changes.\n", + " | \n", + " | Returns:\n", + " | object: Any extra state to store in the module's state_dict\n", + " | \n", + " | get_parameter(self, target: str) -> 'Parameter'\n", + " | Return the parameter given by ``target`` if it exists, otherwise throw an error.\n", + " | \n", + " | See the docstring for ``get_submodule`` for a more detailed\n", + " | explanation of this method's functionality as well as how to\n", + " | correctly specify ``target``.\n", + " | \n", + " | Args:\n", + " | target: The fully-qualified string name of the Parameter\n", + " | to look for. (See ``get_submodule`` for how to specify a\n", + " | fully-qualified string.)\n", + " | \n", + " | Returns:\n", + " | torch.nn.Parameter: The Parameter referenced by ``target``\n", + " | \n", + " | Raises:\n", + " | AttributeError: If the target string references an invalid\n", + " | path or resolves to something that is not an\n", + " | ``nn.Parameter``\n", + " | \n", + " | get_submodule(self, target: str) -> 'Module'\n", + " | Return the submodule given by ``target`` if it exists, otherwise throw an error.\n", + " | \n", + " | For example, let's say you have an ``nn.Module`` ``A`` that\n", + " | looks like this:\n", + " | \n", + " | .. code-block:: text\n", + " | \n", + " | A(\n", + " | (net_b): Module(\n", + " | (net_c): Module(\n", + " | (conv): Conv2d(16, 33, kernel_size=(3, 3), stride=(2, 2))\n", + " | )\n", + " | (linear): Linear(in_features=100, out_features=200, bias=True)\n", + " | )\n", + " | )\n", + " | \n", + " | (The diagram shows an ``nn.Module`` ``A``. ``A`` which has a nested\n", + " | submodule ``net_b``, which itself has two submodules ``net_c``\n", + " | and ``linear``. ``net_c`` then has a submodule ``conv``.)\n", + " | \n", + " | To check whether or not we have the ``linear`` submodule, we\n", + " | would call ``get_submodule(\"net_b.linear\")``. To check whether\n", + " | we have the ``conv`` submodule, we would call\n", + " | ``get_submodule(\"net_b.net_c.conv\")``.\n", + " | \n", + " | The runtime of ``get_submodule`` is bounded by the degree\n", + " | of module nesting in ``target``. A query against\n", + " | ``named_modules`` achieves the same result, but it is O(N) in\n", + " | the number of transitive modules. So, for a simple check to see\n", + " | if some submodule exists, ``get_submodule`` should always be\n", + " | used.\n", + " | \n", + " | Args:\n", + " | target: The fully-qualified string name of the submodule\n", + " | to look for. (See above example for how to specify a\n", + " | fully-qualified string.)\n", + " | \n", + " | Returns:\n", + " | torch.nn.Module: The submodule referenced by ``target``\n", + " | \n", + " | Raises:\n", + " | AttributeError: If at any point along the path resulting from\n", + " | the target string the (sub)path resolves to a non-existent\n", + " | attribute name or an object that is not an instance of ``nn.Module``.\n", + " | \n", + " | half(self) -> Self\n", + " | Casts all floating point parameters and buffers to ``half`` datatype.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | ipu(self, device: Union[torch.device, int, NoneType] = None) -> Self\n", + " | Move all model parameters and buffers to the IPU.\n", + " | \n", + " | This also makes associated parameters and buffers different objects. So\n", + " | it should be called before constructing the optimizer if the module will\n", + " | live on IPU while being optimized.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Arguments:\n", + " | device (int, optional): if specified, all parameters will be\n", + " | copied to that device\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | load_state_dict(self, state_dict: collections.abc.Mapping[str, typing.Any], strict: bool = True, assign: bool = False)\n", + " | Copy parameters and buffers from :attr:`state_dict` into this module and its descendants.\n", + " | \n", + " | If :attr:`strict` is ``True``, then\n", + " | the keys of :attr:`state_dict` must exactly match the keys returned\n", + " | by this module's :meth:`~torch.nn.Module.state_dict` function.\n", + " | \n", + " | .. warning::\n", + " | If :attr:`assign` is ``True`` the optimizer must be created after\n", + " | the call to :attr:`load_state_dict` unless\n", + " | :func:`~torch.__future__.get_swap_module_params_on_conversion` is ``True``.\n", + " | \n", + " | Args:\n", + " | state_dict (dict): a dict containing parameters and\n", + " | persistent buffers.\n", + " | strict (bool, optional): whether to strictly enforce that the keys\n", + " | in :attr:`state_dict` match the keys returned by this module's\n", + " | :meth:`~torch.nn.Module.state_dict` function. Default: ``True``\n", + " | assign (bool, optional): When set to ``False``, the properties of the tensors\n", + " | in the current module are preserved whereas setting it to ``True`` preserves\n", + " | properties of the Tensors in the state dict. The only\n", + " | exception is the ``requires_grad`` field of :class:`~torch.nn.Parameter`\n", + " | for which the value from the module is preserved. Default: ``False``\n", + " | \n", + " | Returns:\n", + " | ``NamedTuple`` with ``missing_keys`` and ``unexpected_keys`` fields:\n", + " | * ``missing_keys`` is a list of str containing any keys that are expected\n", + " | by this module but missing from the provided ``state_dict``.\n", + " | * ``unexpected_keys`` is a list of str containing the keys that are not\n", + " | expected by this module but present in the provided ``state_dict``.\n", + " | \n", + " | Note:\n", + " | If a parameter or buffer is registered as ``None`` and its corresponding key\n", + " | exists in :attr:`state_dict`, :meth:`load_state_dict` will raise a\n", + " | ``RuntimeError``.\n", + " | \n", + " | modules(self) -> collections.abc.Iterator['Module']\n", + " | Return an iterator over all modules in the network.\n", + " | \n", + " | Yields:\n", + " | Module: a module in the network\n", + " | \n", + " | Note:\n", + " | Duplicate modules are returned only once. In the following\n", + " | example, ``l`` will be returned only once.\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> l = nn.Linear(2, 2)\n", + " | >>> net = nn.Sequential(l, l)\n", + " | >>> for idx, m in enumerate(net.modules()):\n", + " | ... print(idx, '->', m)\n", + " | \n", + " | 0 -> Sequential(\n", + " | (0): Linear(in_features=2, out_features=2, bias=True)\n", + " | (1): Linear(in_features=2, out_features=2, bias=True)\n", + " | )\n", + " | 1 -> Linear(in_features=2, out_features=2, bias=True)\n", + " | \n", + " | mtia(self, device: Union[torch.device, int, NoneType] = None) -> Self\n", + " | Move all model parameters and buffers to the MTIA.\n", + " | \n", + " | This also makes associated parameters and buffers different objects. So\n", + " | it should be called before constructing the optimizer if the module will\n", + " | live on MTIA while being optimized.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Arguments:\n", + " | device (int, optional): if specified, all parameters will be\n", + " | copied to that device\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | named_buffers(self, prefix: str = '', recurse: bool = True, remove_duplicate: bool = True) -> collections.abc.Iterator[tuple[str, torch.Tensor]]\n", + " | Return an iterator over module buffers, yielding both the name of the buffer as well as the buffer itself.\n", + " | \n", + " | Args:\n", + " | prefix (str): prefix to prepend to all buffer names.\n", + " | recurse (bool, optional): if True, then yields buffers of this module\n", + " | and all submodules. Otherwise, yields only buffers that\n", + " | are direct members of this module. Defaults to True.\n", + " | remove_duplicate (bool, optional): whether to remove the duplicated buffers in the result. Defaults to True.\n", + " | \n", + " | Yields:\n", + " | (str, torch.Tensor): Tuple containing the name and buffer\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> # xdoctest: +SKIP(\"undefined vars\")\n", + " | >>> for name, buf in self.named_buffers():\n", + " | >>> if name in ['running_var']:\n", + " | >>> print(buf.size())\n", + " | \n", + " | named_children(self) -> collections.abc.Iterator[tuple[str, 'Module']]\n", + " | Return an iterator over immediate children modules, yielding both the name of the module as well as the module itself.\n", + " | \n", + " | Yields:\n", + " | (str, Module): Tuple containing a name and child module\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> # xdoctest: +SKIP(\"undefined vars\")\n", + " | >>> for name, module in model.named_children():\n", + " | >>> if name in ['conv4', 'conv5']:\n", + " | >>> print(module)\n", + " | \n", + " | named_modules(self, memo: Optional[set['Module']] = None, prefix: str = '', remove_duplicate: bool = True)\n", + " | Return an iterator over all modules in the network, yielding both the name of the module as well as the module itself.\n", + " | \n", + " | Args:\n", + " | memo: a memo to store the set of modules already added to the result\n", + " | prefix: a prefix that will be added to the name of the module\n", + " | remove_duplicate: whether to remove the duplicated module instances in the result\n", + " | or not\n", + " | \n", + " | Yields:\n", + " | (str, Module): Tuple of name and module\n", + " | \n", + " | Note:\n", + " | Duplicate modules are returned only once. In the following\n", + " | example, ``l`` will be returned only once.\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> l = nn.Linear(2, 2)\n", + " | >>> net = nn.Sequential(l, l)\n", + " | >>> for idx, m in enumerate(net.named_modules()):\n", + " | ... print(idx, '->', m)\n", + " | \n", + " | 0 -> ('', Sequential(\n", + " | (0): Linear(in_features=2, out_features=2, bias=True)\n", + " | (1): Linear(in_features=2, out_features=2, bias=True)\n", + " | ))\n", + " | 1 -> ('0', Linear(in_features=2, out_features=2, bias=True))\n", + " | \n", + " | named_parameters(self, prefix: str = '', recurse: bool = True, remove_duplicate: bool = True) -> collections.abc.Iterator[tuple[str, torch.nn.parameter.Parameter]]\n", + " | Return an iterator over module parameters, yielding both the name of the parameter as well as the parameter itself.\n", + " | \n", + " | Args:\n", + " | prefix (str): prefix to prepend to all parameter names.\n", + " | recurse (bool): if True, then yields parameters of this module\n", + " | and all submodules. Otherwise, yields only parameters that\n", + " | are direct members of this module.\n", + " | remove_duplicate (bool, optional): whether to remove the duplicated\n", + " | parameters in the result. Defaults to True.\n", + " | \n", + " | Yields:\n", + " | (str, Parameter): Tuple containing the name and parameter\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> # xdoctest: +SKIP(\"undefined vars\")\n", + " | >>> for name, param in self.named_parameters():\n", + " | >>> if name in ['bias']:\n", + " | >>> print(param.size())\n", + " | \n", + " | parameters(self, recurse: bool = True) -> collections.abc.Iterator[torch.nn.parameter.Parameter]\n", + " | Return an iterator over module parameters.\n", + " | \n", + " | This is typically passed to an optimizer.\n", + " | \n", + " | Args:\n", + " | recurse (bool): if True, then yields parameters of this module\n", + " | and all submodules. Otherwise, yields only parameters that\n", + " | are direct members of this module.\n", + " | \n", + " | Yields:\n", + " | Parameter: module parameter\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> # xdoctest: +SKIP(\"undefined vars\")\n", + " | >>> for param in model.parameters():\n", + " | >>> print(type(param), param.size())\n", + " | (20L,)\n", + " | (20L, 1L, 5L, 5L)\n", + " | \n", + " | register_backward_hook(self, hook: Callable[[ForwardRef('Module'), Union[tuple[torch.Tensor, ...], torch.Tensor], Union[tuple[torch.Tensor, ...], torch.Tensor]], Union[NoneType, tuple[torch.Tensor, ...], torch.Tensor]]) -> torch.utils.hooks.RemovableHandle\n", + " | Register a backward hook on the module.\n", + " | \n", + " | This function is deprecated in favor of :meth:`~torch.nn.Module.register_full_backward_hook` and\n", + " | the behavior of this function will change in future versions.\n", + " | \n", + " | Returns:\n", + " | :class:`torch.utils.hooks.RemovableHandle`:\n", + " | a handle that can be used to remove the added hook by calling\n", + " | ``handle.remove()``\n", + " | \n", + " | register_buffer(self, name: str, tensor: Optional[torch.Tensor], persistent: bool = True) -> None\n", + " | Add a buffer to the module.\n", + " | \n", + " | This is typically used to register a buffer that should not be\n", + " | considered a model parameter. For example, BatchNorm's ``running_mean``\n", + " | is not a parameter, but is part of the module's state. Buffers, by\n", + " | default, are persistent and will be saved alongside parameters. This\n", + " | behavior can be changed by setting :attr:`persistent` to ``False``. The\n", + " | only difference between a persistent buffer and a non-persistent buffer\n", + " | is that the latter will not be a part of this module's\n", + " | :attr:`state_dict`.\n", + " | \n", + " | Buffers can be accessed as attributes using given names.\n", + " | \n", + " | Args:\n", + " | name (str): name of the buffer. The buffer can be accessed\n", + " | from this module using the given name\n", + " | tensor (Tensor or None): buffer to be registered. If ``None``, then operations\n", + " | that run on buffers, such as :attr:`cuda`, are ignored. If ``None``,\n", + " | the buffer is **not** included in the module's :attr:`state_dict`.\n", + " | persistent (bool): whether the buffer is part of this module's\n", + " | :attr:`state_dict`.\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> # xdoctest: +SKIP(\"undefined vars\")\n", + " | >>> self.register_buffer('running_mean', torch.zeros(num_features))\n", + " | \n", + " | register_forward_hook(self, hook: Union[Callable[[~T, tuple[Any, ...], Any], Optional[Any]], Callable[[~T, tuple[Any, ...], dict[str, Any], Any], Optional[Any]]], *, prepend: bool = False, with_kwargs: bool = False, always_call: bool = False) -> torch.utils.hooks.RemovableHandle\n", + " | Register a forward hook on the module.\n", + " | \n", + " | The hook will be called every time after :func:`forward` has computed an output.\n", + " | \n", + " | If ``with_kwargs`` is ``False`` or not specified, the input contains only\n", + " | the positional arguments given to the module. Keyword arguments won't be\n", + " | passed to the hooks and only to the ``forward``. The hook can modify the\n", + " | output. It can modify the input inplace but it will not have effect on\n", + " | forward since this is called after :func:`forward` is called. The hook\n", + " | should have the following signature::\n", + " | \n", + " | hook(module, args, output) -> None or modified output\n", + " | \n", + " | If ``with_kwargs`` is ``True``, the forward hook will be passed the\n", + " | ``kwargs`` given to the forward function and be expected to return the\n", + " | output possibly modified. The hook should have the following signature::\n", + " | \n", + " | hook(module, args, kwargs, output) -> None or modified output\n", + " | \n", + " | Args:\n", + " | hook (Callable): The user defined hook to be registered.\n", + " | prepend (bool): If ``True``, the provided ``hook`` will be fired\n", + " | before all existing ``forward`` hooks on this\n", + " | :class:`torch.nn.Module`. Otherwise, the provided\n", + " | ``hook`` will be fired after all existing ``forward`` hooks on\n", + " | this :class:`torch.nn.Module`. Note that global\n", + " | ``forward`` hooks registered with\n", + " | :func:`register_module_forward_hook` will fire before all hooks\n", + " | registered by this method.\n", + " | Default: ``False``\n", + " | with_kwargs (bool): If ``True``, the ``hook`` will be passed the\n", + " | kwargs given to the forward function.\n", + " | Default: ``False``\n", + " | always_call (bool): If ``True`` the ``hook`` will be run regardless of\n", + " | whether an exception is raised while calling the Module.\n", + " | Default: ``False``\n", + " | \n", + " | Returns:\n", + " | :class:`torch.utils.hooks.RemovableHandle`:\n", + " | a handle that can be used to remove the added hook by calling\n", + " | ``handle.remove()``\n", + " | \n", + " | register_forward_pre_hook(self, hook: Union[Callable[[~T, tuple[Any, ...]], Optional[Any]], Callable[[~T, tuple[Any, ...], dict[str, Any]], Optional[tuple[Any, dict[str, Any]]]]], *, prepend: bool = False, with_kwargs: bool = False) -> torch.utils.hooks.RemovableHandle\n", + " | Register a forward pre-hook on the module.\n", + " | \n", + " | The hook will be called every time before :func:`forward` is invoked.\n", + " | \n", + " | \n", + " | If ``with_kwargs`` is false or not specified, the input contains only\n", + " | the positional arguments given to the module. Keyword arguments won't be\n", + " | passed to the hooks and only to the ``forward``. The hook can modify the\n", + " | input. User can either return a tuple or a single modified value in the\n", + " | hook. We will wrap the value into a tuple if a single value is returned\n", + " | (unless that value is already a tuple). The hook should have the\n", + " | following signature::\n", + " | \n", + " | hook(module, args) -> None or modified input\n", + " | \n", + " | If ``with_kwargs`` is true, the forward pre-hook will be passed the\n", + " | kwargs given to the forward function. And if the hook modifies the\n", + " | input, both the args and kwargs should be returned. The hook should have\n", + " | the following signature::\n", + " | \n", + " | hook(module, args, kwargs) -> None or a tuple of modified input and kwargs\n", + " | \n", + " | Args:\n", + " | hook (Callable): The user defined hook to be registered.\n", + " | prepend (bool): If true, the provided ``hook`` will be fired before\n", + " | all existing ``forward_pre`` hooks on this\n", + " | :class:`torch.nn.Module`. Otherwise, the provided\n", + " | ``hook`` will be fired after all existing ``forward_pre`` hooks\n", + " | on this :class:`torch.nn.Module`. Note that global\n", + " | ``forward_pre`` hooks registered with\n", + " | :func:`register_module_forward_pre_hook` will fire before all\n", + " | hooks registered by this method.\n", + " | Default: ``False``\n", + " | with_kwargs (bool): If true, the ``hook`` will be passed the kwargs\n", + " | given to the forward function.\n", + " | Default: ``False``\n", + " | \n", + " | Returns:\n", + " | :class:`torch.utils.hooks.RemovableHandle`:\n", + " | a handle that can be used to remove the added hook by calling\n", + " | ``handle.remove()``\n", + " | \n", + " | register_full_backward_hook(self, hook: Callable[[ForwardRef('Module'), Union[tuple[torch.Tensor, ...], torch.Tensor], Union[tuple[torch.Tensor, ...], torch.Tensor]], Union[NoneType, tuple[torch.Tensor, ...], torch.Tensor]], prepend: bool = False) -> torch.utils.hooks.RemovableHandle\n", + " | Register a backward hook on the module.\n", + " | \n", + " | The hook will be called every time the gradients with respect to a module are computed, and its firing rules are as follows:\n", + " | \n", + " | 1. Ordinarily, the hook fires when the gradients are computed with respect to the module inputs.\n", + " | 2. If none of the module inputs require gradients, the hook will fire when the gradients are computed\n", + " | with respect to module outputs.\n", + " | 3. If none of the module outputs require gradients, then the hooks will not fire.\n", + " | \n", + " | The hook should have the following signature::\n", + " | \n", + " | hook(module, grad_input, grad_output) -> tuple(Tensor) or None\n", + " | \n", + " | The :attr:`grad_input` and :attr:`grad_output` are tuples that contain the gradients\n", + " | with respect to the inputs and outputs respectively. The hook should\n", + " | not modify its arguments, but it can optionally return a new gradient with\n", + " | respect to the input that will be used in place of :attr:`grad_input` in\n", + " | subsequent computations. :attr:`grad_input` will only correspond to the inputs given\n", + " | as positional arguments and all kwarg arguments are ignored. Entries\n", + " | in :attr:`grad_input` and :attr:`grad_output` will be ``None`` for all non-Tensor\n", + " | arguments.\n", + " | \n", + " | For technical reasons, when this hook is applied to a Module, its forward function will\n", + " | receive a view of each Tensor passed to the Module. Similarly the caller will receive a view\n", + " | of each Tensor returned by the Module's forward function.\n", + " | \n", + " | .. warning ::\n", + " | Modifying inputs or outputs inplace is not allowed when using backward hooks and\n", + " | will raise an error.\n", + " | \n", + " | Args:\n", + " | hook (Callable): The user-defined hook to be registered.\n", + " | prepend (bool): If true, the provided ``hook`` will be fired before\n", + " | all existing ``backward`` hooks on this\n", + " | :class:`torch.nn.Module`. Otherwise, the provided\n", + " | ``hook`` will be fired after all existing ``backward`` hooks on\n", + " | this :class:`torch.nn.Module`. Note that global\n", + " | ``backward`` hooks registered with\n", + " | :func:`register_module_full_backward_hook` will fire before\n", + " | all hooks registered by this method.\n", + " | \n", + " | Returns:\n", + " | :class:`torch.utils.hooks.RemovableHandle`:\n", + " | a handle that can be used to remove the added hook by calling\n", + " | ``handle.remove()``\n", + " | \n", + " | register_full_backward_pre_hook(self, hook: Callable[[ForwardRef('Module'), Union[tuple[torch.Tensor, ...], torch.Tensor]], Union[NoneType, tuple[torch.Tensor, ...], torch.Tensor]], prepend: bool = False) -> torch.utils.hooks.RemovableHandle\n", + " | Register a backward pre-hook on the module.\n", + " | \n", + " | The hook will be called every time the gradients for the module are computed.\n", + " | The hook should have the following signature::\n", + " | \n", + " | hook(module, grad_output) -> tuple[Tensor] or None\n", + " | \n", + " | The :attr:`grad_output` is a tuple. The hook should\n", + " | not modify its arguments, but it can optionally return a new gradient with\n", + " | respect to the output that will be used in place of :attr:`grad_output` in\n", + " | subsequent computations. Entries in :attr:`grad_output` will be ``None`` for\n", + " | all non-Tensor arguments.\n", + " | \n", + " | For technical reasons, when this hook is applied to a Module, its forward function will\n", + " | receive a view of each Tensor passed to the Module. Similarly the caller will receive a view\n", + " | of each Tensor returned by the Module's forward function.\n", + " | \n", + " | .. warning ::\n", + " | Modifying inputs inplace is not allowed when using backward hooks and\n", + " | will raise an error.\n", + " | \n", + " | Args:\n", + " | hook (Callable): The user-defined hook to be registered.\n", + " | prepend (bool): If true, the provided ``hook`` will be fired before\n", + " | all existing ``backward_pre`` hooks on this\n", + " | :class:`torch.nn.Module`. Otherwise, the provided\n", + " | ``hook`` will be fired after all existing ``backward_pre`` hooks\n", + " | on this :class:`torch.nn.Module`. Note that global\n", + " | ``backward_pre`` hooks registered with\n", + " | :func:`register_module_full_backward_pre_hook` will fire before\n", + " | all hooks registered by this method.\n", + " | \n", + " | Returns:\n", + " | :class:`torch.utils.hooks.RemovableHandle`:\n", + " | a handle that can be used to remove the added hook by calling\n", + " | ``handle.remove()``\n", + " | \n", + " | register_load_state_dict_post_hook(self, hook)\n", + " | Register a post-hook to be run after module's :meth:`~nn.Module.load_state_dict` is called.\n", + " | \n", + " | It should have the following signature::\n", + " | hook(module, incompatible_keys) -> None\n", + " | \n", + " | The ``module`` argument is the current module that this hook is registered\n", + " | on, and the ``incompatible_keys`` argument is a ``NamedTuple`` consisting\n", + " | of attributes ``missing_keys`` and ``unexpected_keys``. ``missing_keys``\n", + " | is a ``list`` of ``str`` containing the missing keys and\n", + " | ``unexpected_keys`` is a ``list`` of ``str`` containing the unexpected keys.\n", + " | \n", + " | The given incompatible_keys can be modified inplace if needed.\n", + " | \n", + " | Note that the checks performed when calling :func:`load_state_dict` with\n", + " | ``strict=True`` are affected by modifications the hook makes to\n", + " | ``missing_keys`` or ``unexpected_keys``, as expected. Additions to either\n", + " | set of keys will result in an error being thrown when ``strict=True``, and\n", + " | clearing out both missing and unexpected keys will avoid an error.\n", + " | \n", + " | Returns:\n", + " | :class:`torch.utils.hooks.RemovableHandle`:\n", + " | a handle that can be used to remove the added hook by calling\n", + " | ``handle.remove()``\n", + " | \n", + " | register_load_state_dict_pre_hook(self, hook)\n", + " | Register a pre-hook to be run before module's :meth:`~nn.Module.load_state_dict` is called.\n", + " | \n", + " | It should have the following signature::\n", + " | hook(module, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs) -> None # noqa: B950\n", + " | \n", + " | Arguments:\n", + " | hook (Callable): Callable hook that will be invoked before\n", + " | loading the state dict.\n", + " | \n", + " | register_module(self, name: str, module: Optional[ForwardRef('Module')]) -> None\n", + " | Alias for :func:`add_module`.\n", + " | \n", + " | register_parameter(self, name: str, param: Optional[torch.nn.parameter.Parameter]) -> None\n", + " | Add a parameter to the module.\n", + " | \n", + " | The parameter can be accessed as an attribute using given name.\n", + " | \n", + " | Args:\n", + " | name (str): name of the parameter. The parameter can be accessed\n", + " | from this module using the given name\n", + " | param (Parameter or None): parameter to be added to the module. If\n", + " | ``None``, then operations that run on parameters, such as :attr:`cuda`,\n", + " | are ignored. If ``None``, the parameter is **not** included in the\n", + " | module's :attr:`state_dict`.\n", + " | \n", + " | register_state_dict_post_hook(self, hook)\n", + " | Register a post-hook for the :meth:`~torch.nn.Module.state_dict` method.\n", + " | \n", + " | It should have the following signature::\n", + " | hook(module, state_dict, prefix, local_metadata) -> None\n", + " | \n", + " | The registered hooks can modify the ``state_dict`` inplace.\n", + " | \n", + " | register_state_dict_pre_hook(self, hook)\n", + " | Register a pre-hook for the :meth:`~torch.nn.Module.state_dict` method.\n", + " | \n", + " | It should have the following signature::\n", + " | hook(module, prefix, keep_vars) -> None\n", + " | \n", + " | The registered hooks can be used to perform pre-processing before the ``state_dict``\n", + " | call is made.\n", + " | \n", + " | requires_grad_(self, requires_grad: bool = True) -> Self\n", + " | Change if autograd should record operations on parameters in this module.\n", + " | \n", + " | This method sets the parameters' :attr:`requires_grad` attributes\n", + " | in-place.\n", + " | \n", + " | This method is helpful for freezing part of the module for finetuning\n", + " | or training parts of a model individually (e.g., GAN training).\n", + " | \n", + " | See :ref:`locally-disable-grad-doc` for a comparison between\n", + " | `.requires_grad_()` and several similar mechanisms that may be confused with it.\n", + " | \n", + " | Args:\n", + " | requires_grad (bool): whether autograd should record operations on\n", + " | parameters in this module. Default: ``True``.\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | set_extra_state(self, state: Any) -> None\n", + " | Set extra state contained in the loaded `state_dict`.\n", + " | \n", + " | This function is called from :func:`load_state_dict` to handle any extra state\n", + " | found within the `state_dict`. Implement this function and a corresponding\n", + " | :func:`get_extra_state` for your module if you need to store extra state within its\n", + " | `state_dict`.\n", + " | \n", + " | Args:\n", + " | state (dict): Extra state from the `state_dict`\n", + " | \n", + " | set_submodule(self, target: str, module: 'Module', strict: bool = False) -> None\n", + " | Set the submodule given by ``target`` if it exists, otherwise throw an error.\n", + " | \n", + " | .. note::\n", + " | If ``strict`` is set to ``False`` (default), the method will replace an existing submodule\n", + " | or create a new submodule if the parent module exists. If ``strict`` is set to ``True``,\n", + " | the method will only attempt to replace an existing submodule and throw an error if\n", + " | the submodule does not exist.\n", + " | \n", + " | For example, let's say you have an ``nn.Module`` ``A`` that\n", + " | looks like this:\n", + " | \n", + " | .. code-block:: text\n", + " | \n", + " | A(\n", + " | (net_b): Module(\n", + " | (net_c): Module(\n", + " | (conv): Conv2d(3, 3, 3)\n", + " | )\n", + " | (linear): Linear(3, 3)\n", + " | )\n", + " | )\n", + " | \n", + " | (The diagram shows an ``nn.Module`` ``A``. ``A`` has a nested\n", + " | submodule ``net_b``, which itself has two submodules ``net_c``\n", + " | and ``linear``. ``net_c`` then has a submodule ``conv``.)\n", + " | \n", + " | To override the ``Conv2d`` with a new submodule ``Linear``, you\n", + " | could call ``set_submodule(\"net_b.net_c.conv\", nn.Linear(1, 1))``\n", + " | where ``strict`` could be ``True`` or ``False``\n", + " | \n", + " | To add a new submodule ``Conv2d`` to the existing ``net_b`` module,\n", + " | you would call ``set_submodule(\"net_b.conv\", nn.Conv2d(1, 1, 1))``.\n", + " | \n", + " | In the above if you set ``strict=True`` and call\n", + " | ``set_submodule(\"net_b.conv\", nn.Conv2d(1, 1, 1), strict=True)``, an AttributeError\n", + " | will be raised because ``net_b`` does not have a submodule named ``conv``.\n", + " | \n", + " | Args:\n", + " | target: The fully-qualified string name of the submodule\n", + " | to look for. (See above example for how to specify a\n", + " | fully-qualified string.)\n", + " | module: The module to set the submodule to.\n", + " | strict: If ``False``, the method will replace an existing submodule\n", + " | or create a new submodule if the parent module exists. If ``True``,\n", + " | the method will only attempt to replace an existing submodule and throw an error\n", + " | if the submodule doesn't already exist.\n", + " | \n", + " | Raises:\n", + " | ValueError: If the ``target`` string is empty or if ``module`` is not an instance of ``nn.Module``.\n", + " | AttributeError: If at any point along the path resulting from\n", + " | the ``target`` string the (sub)path resolves to a non-existent\n", + " | attribute name or an object that is not an instance of ``nn.Module``.\n", + " | \n", + " | share_memory(self) -> Self\n", + " | See :meth:`torch.Tensor.share_memory_`.\n", + " | \n", + " | state_dict(self, *args, destination=None, prefix='', keep_vars=False)\n", + " | Return a dictionary containing references to the whole state of the module.\n", + " | \n", + " | Both parameters and persistent buffers (e.g. running averages) are\n", + " | included. Keys are corresponding parameter and buffer names.\n", + " | Parameters and buffers set to ``None`` are not included.\n", + " | \n", + " | .. note::\n", + " | The returned object is a shallow copy. It contains references\n", + " | to the module's parameters and buffers.\n", + " | \n", + " | .. warning::\n", + " | Currently ``state_dict()`` also accepts positional arguments for\n", + " | ``destination``, ``prefix`` and ``keep_vars`` in order. However,\n", + " | this is being deprecated and keyword arguments will be enforced in\n", + " | future releases.\n", + " | \n", + " | .. warning::\n", + " | Please avoid the use of argument ``destination`` as it is not\n", + " | designed for end-users.\n", + " | \n", + " | Args:\n", + " | destination (dict, optional): If provided, the state of module will\n", + " | be updated into the dict and the same object is returned.\n", + " | Otherwise, an ``OrderedDict`` will be created and returned.\n", + " | Default: ``None``.\n", + " | prefix (str, optional): a prefix added to parameter and buffer\n", + " | names to compose the keys in state_dict. Default: ``''``.\n", + " | keep_vars (bool, optional): by default the :class:`~torch.Tensor` s\n", + " | returned in the state dict are detached from autograd. If it's\n", + " | set to ``True``, detaching will not be performed.\n", + " | Default: ``False``.\n", + " | \n", + " | Returns:\n", + " | dict:\n", + " | a dictionary containing a whole state of the module\n", + " | \n", + " | Example::\n", + " | \n", + " | >>> # xdoctest: +SKIP(\"undefined vars\")\n", + " | >>> module.state_dict().keys()\n", + " | ['bias', 'weight']\n", + " | \n", + " | to(self, *args, **kwargs)\n", + " | Move and/or cast the parameters and buffers.\n", + " | \n", + " | This can be called as\n", + " | \n", + " | .. function:: to(device=None, dtype=None, non_blocking=False)\n", + " | :noindex:\n", + " | \n", + " | .. function:: to(dtype, non_blocking=False)\n", + " | :noindex:\n", + " | \n", + " | .. function:: to(tensor, non_blocking=False)\n", + " | :noindex:\n", + " | \n", + " | .. function:: to(memory_format=torch.channels_last)\n", + " | :noindex:\n", + " | \n", + " | Its signature is similar to :meth:`torch.Tensor.to`, but only accepts\n", + " | floating point or complex :attr:`dtype`\\ s. In addition, this method will\n", + " | only cast the floating point or complex parameters and buffers to :attr:`dtype`\n", + " | (if given). The integral parameters and buffers will be moved\n", + " | :attr:`device`, if that is given, but with dtypes unchanged. When\n", + " | :attr:`non_blocking` is set, it tries to convert/move asynchronously\n", + " | with respect to the host if possible, e.g., moving CPU Tensors with\n", + " | pinned memory to CUDA devices.\n", + " | \n", + " | See below for examples.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Args:\n", + " | device (:class:`torch.device`): the desired device of the parameters\n", + " | and buffers in this module\n", + " | dtype (:class:`torch.dtype`): the desired floating point or complex dtype of\n", + " | the parameters and buffers in this module\n", + " | tensor (torch.Tensor): Tensor whose dtype and device are the desired\n", + " | dtype and device for all parameters and buffers in this module\n", + " | memory_format (:class:`torch.memory_format`): the desired memory\n", + " | format for 4D parameters and buffers in this module (keyword\n", + " | only argument)\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | Examples::\n", + " | \n", + " | >>> # xdoctest: +IGNORE_WANT(\"non-deterministic\")\n", + " | >>> linear = nn.Linear(2, 2)\n", + " | >>> linear.weight\n", + " | Parameter containing:\n", + " | tensor([[ 0.1913, -0.3420],\n", + " | [-0.5113, -0.2325]])\n", + " | >>> linear.to(torch.double)\n", + " | Linear(in_features=2, out_features=2, bias=True)\n", + " | >>> linear.weight\n", + " | Parameter containing:\n", + " | tensor([[ 0.1913, -0.3420],\n", + " | [-0.5113, -0.2325]], dtype=torch.float64)\n", + " | >>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_CUDA1)\n", + " | >>> gpu1 = torch.device(\"cuda:1\")\n", + " | >>> linear.to(gpu1, dtype=torch.half, non_blocking=True)\n", + " | Linear(in_features=2, out_features=2, bias=True)\n", + " | >>> linear.weight\n", + " | Parameter containing:\n", + " | tensor([[ 0.1914, -0.3420],\n", + " | [-0.5112, -0.2324]], dtype=torch.float16, device='cuda:1')\n", + " | >>> cpu = torch.device(\"cpu\")\n", + " | >>> linear.to(cpu)\n", + " | Linear(in_features=2, out_features=2, bias=True)\n", + " | >>> linear.weight\n", + " | Parameter containing:\n", + " | tensor([[ 0.1914, -0.3420],\n", + " | [-0.5112, -0.2324]], dtype=torch.float16)\n", + " | \n", + " | >>> linear = nn.Linear(2, 2, bias=None).to(torch.cdouble)\n", + " | >>> linear.weight\n", + " | Parameter containing:\n", + " | tensor([[ 0.3741+0.j, 0.2382+0.j],\n", + " | [ 0.5593+0.j, -0.4443+0.j]], dtype=torch.complex128)\n", + " | >>> linear(torch.ones(3, 2, dtype=torch.cdouble))\n", + " | tensor([[0.6122+0.j, 0.1150+0.j],\n", + " | [0.6122+0.j, 0.1150+0.j],\n", + " | [0.6122+0.j, 0.1150+0.j]], dtype=torch.complex128)\n", + " | \n", + " | to_empty(self, *, device: Union[int, str, torch.device, NoneType], recurse: bool = True) -> Self\n", + " | Move the parameters and buffers to the specified device without copying storage.\n", + " | \n", + " | Args:\n", + " | device (:class:`torch.device`): The desired device of the parameters\n", + " | and buffers in this module.\n", + " | recurse (bool): Whether parameters and buffers of submodules should\n", + " | be recursively moved to the specified device.\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | train(self, mode: bool = True) -> Self\n", + " | Set the module in training mode.\n", + " | \n", + " | This has an effect only on certain modules. See the documentation of\n", + " | particular modules for details of their behaviors in training/evaluation\n", + " | mode, i.e., whether they are affected, e.g. :class:`Dropout`, :class:`BatchNorm`,\n", + " | etc.\n", + " | \n", + " | Args:\n", + " | mode (bool): whether to set training mode (``True``) or evaluation\n", + " | mode (``False``). Default: ``True``.\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | type(self, dst_type: Union[torch.dtype, str]) -> Self\n", + " | Casts all parameters and buffers to :attr:`dst_type`.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Args:\n", + " | dst_type (type or string): the desired type\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | xpu(self, device: Union[torch.device, int, NoneType] = None) -> Self\n", + " | Move all model parameters and buffers to the XPU.\n", + " | \n", + " | This also makes associated parameters and buffers different objects. So\n", + " | it should be called before constructing optimizer if the module will\n", + " | live on XPU while being optimized.\n", + " | \n", + " | .. note::\n", + " | This method modifies the module in-place.\n", + " | \n", + " | Arguments:\n", + " | device (int, optional): if specified, all parameters will be\n", + " | copied to that device\n", + " | \n", + " | Returns:\n", + " | Module: self\n", + " | \n", + " | zero_grad(self, set_to_none: bool = True) -> None\n", + " | Reset gradients of all model parameters.\n", + " | \n", + " | See similar function under :class:`torch.optim.Optimizer` for more context.\n", + " | \n", + " | Args:\n", + " | set_to_none (bool): instead of setting to zero, set the grads to None.\n", + " | See :meth:`torch.optim.Optimizer.zero_grad` for details.\n", + " | \n", + " | ----------------------------------------------------------------------\n", + " | Data descriptors inherited from torch.nn.modules.module.Module:\n", + " | \n", + " | __dict__\n", + " | dictionary for instance variables (if defined)\n", + " | \n", + " | __weakref__\n", + " | list of weak references to the object (if defined)\n", + " | \n", + " | ----------------------------------------------------------------------\n", + " | Data and other attributes inherited from torch.nn.modules.module.Module:\n", + " | \n", + " | T_destination = ~T_destination\n", + " | \n", + " | call_super_init = False\n", + " | \n", + " | dump_patches = False\n", + "\n" + ] + } + ], + "source": [ + "help(metaABaCo)" + ] + }, + { + "cell_type": "markdown", + "id": "a6f00887", + "metadata": {}, + "source": [ + "### Training the ABaCo model\n", + "\n", + "To train ABaCo on the prepared dataset, we then use method `abaco.metaABaCo.fit()`. " + ] + }, + { + "cell_type": "code", + "execution_count": 266, + "id": "e5c0aec5", + "metadata": { + "lines_to_next_cell": 2, + "tags": [ + "hide-output" + ] + }, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "Training: VAE for learning meaningful embeddings: 100%|██████████| 3000/3000 [00:40<00:00, 74.64it/s, bio_penalty=0.0000, clustering_loss=0.0000, elbo=257.3351, epoch=2999/3001, vae_loss=257.3351]\n", + "Training: Embeddings batch effect correction using adversrial training: 100%|██████████| 1000/1000 [00:17<00:00, 56.72it/s, adv_loss=-0.5191, bio_penalty=0.0000, clustering_loss=0.0000, disc_loss=0.5191, elbo=281.0243, epoch=999/1001, vae_loss=281.0243]\n", + "Training: VAE decoder with masked batch labels: 100%|██████████| 3000/3000 [00:27<00:00, 108.09it/s, cycle_loss=0.0000, epoch=3000/3000, vae_loss=326.0634]\n" + ] + } + ], + "source": [ + "# Train the model,\n", + "model.fit(\n", + " seed=42,\n", + " w_elbo_nll=10, # more emphasis on reconstruction\n", + " w_bio_penalty=0.0, # disable bio supervision\n", + " w_cluster_penalty=0.0, # light regularization or can be 0\n", + " w_disc=1.0, # default, discriminator training strength\n", + ")" + ] + }, + { + "cell_type": "markdown", + "id": "f02f6b0d", + "metadata": {}, + "source": [ + "`abaco.metaABaCo` provides a method that visualizes the latent vectors via PCA for dimensionality reduction: `.plot_pca_posterior()`" + ] + }, + { + "cell_type": "code", + "execution_count": 268, + "id": "58848ce7", + "metadata": {}, + "outputs": [ + { + "data": { + "image/png": "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", + "text/plain": [ + "
" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "model.plot_pca_posterior()" + ] + }, + { + "cell_type": "markdown", + "id": "18852212", + "metadata": {}, + "source": [ + "### Reconstructing the dataset with ABaCo\n", + "\n", + "To reconstruct the dataset we use method `abaco.metaABaCo.correct()` and save it as a new layer `abaco` in the annotated dataset. " + ] + }, + { + "cell_type": "code", + "execution_count": 269, + "id": "2c9dbf2a", + "metadata": {}, + "outputs": [], + "source": [ + "# Reconstruct the dataset using the trained ABaCo model\n", + "corrected_dataset = model.correct(seed=42)\n", + "# save back to adata\n", + "adata.layers['abaco'] = corrected_dataset.set_index(id_col).drop(columns=[batch_col, bio_col]).values\n", + "adata.write_h5ad(\"data/mgnify_tomato.h5ad\")" + ] + }, + { + "cell_type": "markdown", + "id": "f0d6c9c1", + "metadata": {}, + "source": [ + "## Viewing the ABaCo reconstructed dataset\n", + "\n", + "Here we again take a look at the PCoA but first lets check out the mean differences between the original and batch corrected counts." + ] + }, + { + "cell_type": "code", + "execution_count": 327, + "id": "b31f4a15", + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "(24288,) (24288,)\n" + ] + }, + { + "data": { + "text/html": [ + "
" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "# flatten \n", + "nona = adata.layers['nona'].flatten()\n", + "abaco = adata.layers['abaco'].flatten()\n", + "# calculate means and diffs\n", + "means = (nona + abaco) / 2\n", + "diffs = nona - abaco\n", + "print(means.shape, diffs.shape)\n", + "# mean and limits of agreement\n", + "mean_diff = np.mean(diffs)\n", + "std_diff = np.std(diffs)\n", + "loa_upper = mean_diff + 1.96 * std_diff\n", + "loa_lower = mean_diff - 1.96 * std_diff\n", + "\n", + "# creating plot\n", + "fig = go.Figure()\n", + "fig.add_trace(\n", + " go.Scatter(\n", + " x=means, y=diffs, \n", + " mode='markers', \n", + " marker=dict(color='blue', opacity=0.1)\n", + " )\n", + ")\n", + "fig.add_hline(y=mean_diff,line_color=\"black\", annotation_text=\"Mean Diff\")\n", + "fig.add_hline(y=loa_upper, line_dash=\"dot\", line_color=\"black\")\n", + "fig.add_hline(y=loa_lower, line_dash=\"dot\", line_color=\"black\")\n", + "# titles\n", + "fig.update_layout(\n", + " template=\"plotly_white\",\n", + " title=\"Bland-Altman Plot: Original vs. ABaCo Corrected\",\n", + " xaxis_title=\"Mean of Original and ABaCo Corrected\",\n", + " yaxis_title=\"Difference (original - abaco)\",\n", + ")\n", + "fig.show()" + ] + }, + { + "cell_type": "markdown", + "id": "f39389fb", + "metadata": {}, + "source": [ + "No clear tendency for abaco correction to be lower or higher than original counts." + ] + }, + { + "cell_type": "code", + "execution_count": 329, + "id": "a23a2c21", + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/Users/anglup/GitHub/abaco/.venv/lib/python3.11/site-packages/skbio/stats/ordination/_principal_coordinate_analysis.py:157: RuntimeWarning:\n", + "\n", + "EIGH: since no value for dimensions is specified, PCoA for all dimensions will be computed, which may result in long computation time if the original distance matrix is large.\n", + "\n" + ] + }, + { + "data": { + "text/html": [ + "
" + ] + }, + "metadata": {}, + "output_type": "display_data" + } + ], + "source": [ + "# Plot the PCoA of the reconstructed dataset\n", + "plotPCoA(\n", + " data = corrected_dataset, \n", + " sample_label=id_col, \n", + " batch_label=batch_col, \n", + " experiment_label=bio_col\n", + ")" + ] + }, + { + "cell_type": "markdown", + "id": "1e761983", + "metadata": {}, + "source": [ + "In the PCoA we see that there is more overlap of the batches based on the colours mixing. " + ] + }, + { + "cell_type": "markdown", + "id": "b433f7df", + "metadata": {}, + "source": [ + "## Conclusion\n", + "\n", + "The goal was to: \n", + "\n", + "✅ correct the batch effect (reduce clustering of points by colour in the PCoA and pairplot coloured by batches)\n", + "\n", + "A brief visual inspection of the PCoA suggests that ABaCo reduced the batch effect associated with analyzing the samples from different instruments, while still maintaining the original biological variance.\n", + "\n", + "\n", + "---" + ] + }, + { + "cell_type": "markdown", + "id": "2ebec2c5", + "metadata": {}, + "source": [] + } + ], + "metadata": { + "kernelspec": { + "display_name": ".venv", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.11.0" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} diff --git a/docs/tutorial/demo-mgnify-tomatoes.py b/docs/tutorial/demo-mgnify-tomatoes.py new file mode 100644 index 0000000..7186fb6 --- /dev/null +++ b/docs/tutorial/demo-mgnify-tomatoes.py @@ -0,0 +1,445 @@ +# --- +# jupyter: +# jupytext: +# text_representation: +# extension: .py +# format_name: percent +# format_version: '1.3' +# jupytext_version: 1.18.1 +# kernelspec: +# display_name: .venv +# language: python +# name: python3 +# --- + +# %% [markdown] +# # `ABaCo` demo: No biological labels w/ Mgnify tomatoes + +# %% [markdown] +# In this tutorial we demonstrate how ABaCo can be used: +# - without known biological group labels (e.g., only correct batch effect) +# - and for integrating samples from multiple studies. +# +# Specifically, we integrate tomato rhizosphere metagenomics analyses available on MGniFy. +# - MGYS00006231, Illumina HiSeq 4000 +# - MGYS00006204, Illumina HiSeq 2500 +# - MGYS00006205, Illumina HiSeq 2500 +# +# **Additionally** we provide an intro to [working with anndata](https://anndata.readthedocs.io/en/stable/tutorials/notebooks/getting-started.html) +# +# **Note** in this tutorial we include a dummy bio column for demo purposes, but it is not necessary to have a bio column to use abaco. +# +# ----- +# **Goal:** +# +# With ABaCo the aim is to remove the technical variation between sequencing platforms. +# +# ----- +# +# To start we retrieve the data from MGniFy. + +# %% [markdown] +# ## Downloading the data + +# %% +import requests +from io import StringIO +import pandas as pd +import numpy as np + +# urls manually retrieved from mgnify website +datasets = { + "MGYS00006205": "https://www.ebi.ac.uk/metagenomics/api/v1/studies/MGYS00006205/pipelines/5.0/file/ERP140107_taxonomy_abundances_SSU_v5.0.tsv", + "MGYS00006231": "https://www.ebi.ac.uk/metagenomics/api/v1/studies/MGYS00006231/pipelines/5.0/file/ERP139927_taxonomy_abundances_SSU_v5.0.tsv", + "MGYS00006204": "https://www.ebi.ac.uk/metagenomics/api/v1/studies/MGYS00006204/pipelines/5.0/file/ERP140102_taxonomy_abundances_SSU_v5.0.tsv", +} +# init for dfs +dfs = {} +for key, url in datasets.items(): + # download + response = requests.get(url) + # treat as file + file_like = StringIO(response.content.decode('utf-8')) + # read into pd df + dfs[key] = pd.read_csv(file_like, sep='\t') + # rename sampleid col to taxa + dfs[key] = dfs[key].rename(columns={"#SampleID": "taxa"}) + # set index to taxa + dfs[key] = dfs[key].set_index('taxa') + # drop rows with all zeros + dfs[key] = dfs[key][dfs[key].sum(axis=1)>0] + # sanity check + print(dfs[key].shape) + display(dfs[key].sample(5)) + +# %% [markdown] +# Based on the sampled outputs printed above, w can see that the data: +# - has taxa as rows +# - the samples as columns +# - and the values are the number of taxanomic assignments for the given taxa x sample +# +# To join them we will use a helper function `abaco.utils.df_joiner()`. + +# %% +from abaco.utils import df_joiner + +joined = df_joiner( + df_dict=dfs, + on="taxa", + how="outer", +) + +joined.info() + +# %% [markdown] +# ## A brief intro to `anndata` +# +# ![img](https://raw.githubusercontent.com/scverse/anndata/main/docs/_static/img/anndata_schema.svg) +# +# For this example we will use anndata to store multiple layers of the data (e.g., raw counts, normalized, etc.) and annotate the samples and taxa with metadata. Below we prepare the annotations (i.e., obs, var). + +# %% +import anndata as ad +import scanpy as sc + +# preparing obs metadata +dfs_T = {} +# for each study +for key in dfs: + # transpose + dfs_T[key] = dfs[key].T + # add study source column to be batch effect + dfs_T[key]['source'] = key + # dummy bio column, rando group assign, for checking and demo + dfs_T[key]['bio'] = np.random.choice(["Group1", "Group2"], size = len(dfs_T[key])) + # drop allother cols + dfs_T[key] = dfs_T[key][['source', 'bio']] +# now concat all study metadata +obs = pd.concat(dfs_T.values()) +print(obs.shape) +display(obs.sample(5)) + +# preparing var metadata (taxa levels) +# get taxa index as df and split on ';' +var = joined.reset_index()[['taxa']] +var = var['taxa'].str.split(';', expand=True) +# rename cols +var.columns = [ + "superkingdom", "kingdom","phylum", "class", + "order", "family", "genus", "species" +] +print(var.shape) +display(var.sample(5)) + +# %% [markdown] +# We can initiate the AnnData objct with the values (taxanomic assignment counts) and the metadata we prepped above. + +# %% +# init anndata object +full = ad.AnnData( + X=joined.T.values, + obs=obs, + var=var, +) +# check it out +full + +# %% [markdown] +# With anndata we can also store additional layers seamlessly with the same annoations. Below we will create a layer with the NaNs filled with 0s + +# %% +# layer with no na +full.layers['nona'] = np.nan_to_num(full.X, nan=0) +print(full) + +# checking it out, also demoing to_df method +print("original:") +display(full.to_df().head(5)) +print("no na:") +display(full.to_df(layer='nona').head(5)) + +# %% [markdown] +# we can also add normalized and clr layers. For more details on the reasoning of these layers visit our other tutorials. + +# %% +from skbio.stats.composition import clr + +full.layers['norm'] = full.layers['nona']/full.layers['nona'].sum(axis=1).reshape(-1,1) + +# CLR transform normalized data, replacing zeros with smallll val to avoid log(0) +full.layers['clr'] = clr(np.where(full.layers['norm'] > 0, full.layers['norm'], 1e-10)) + +print(full) + +# %% [markdown] +# With anndata we can also filter the data using the annotations/metadata. Below we will keep the variables (taxa) with genus assignment. Additionally we filter out some taxa if they are missing in more than 30% of the samples. + +# %% +# must have genus level +adata = full[:, ~full.var['genus'].isna()] +print(adata) + +# filtering if taxa is missing (NaN) in more than 30% of samples +adata = adata[:, np.isnan(adata.X).sum(axis=0)/adata.X.shape[0] <= 0.3] +print(adata) + +# %% [markdown] +# We are down to 176 taxa. +# +# However, it can be trickier to aggregate with anndata -- but we can use the [scanpy toolkit](https://scanpy.readthedocs.io/en/stable/index.html) to help. +# +# Specifically we will use [scanpy.get.aggregate()](https://scanpy.readthedocs.io/en/stable/generated/scanpy.get.aggregate.html) to help us check that the dummy biological groups we created are not different. + +# %% +# agg by fake bio group +agg_mean = sc.get.aggregate( + # the anndata object + adata=adata, + # what layer to use + layer='clr', + # group by col + by="bio", + # axis that the group lives + axis="obs", + # how to agg + func='mean', +) +# check out the returned AnnData objct +print(agg_mean) +# we can see taht the default layer name if not provided is the func name +display(agg_mean.to_df('mean').head(5)) + +# %% [markdown] +# Now lets quickly check that group 1 and 2 are not different visually and quick stats. + +# %% +import plotly.graph_objects as go +from scipy import stats + +# transpose for easier plotting +bio_sum = agg_mean.to_df('mean').T + +# init fig +fig = go.Figure() +# add traces +fig.add_trace(go.Histogram(x=bio_sum['Group1'], name='Group1')) +fig.add_trace(go.Histogram(x=bio_sum['Group2'], name='Group2')) + +# formatting +fig.update_layout(barmode='overlay') +fig.update_traces(opacity=0.75) +fig.show() + +# check no diff in stats +stats.wilcoxon(bio_sum['Group1'], bio_sum['Group2']) + +# %% [markdown] +# ## Prepare data for ABaCo +# +# Great. Now lets proceed with adata where we can easily extract a df to meet the data format required for abaco. + +# %% +batch_col = 'source' +bio_col = 'bio' +id_col = 'index' + +# the raw counts as df +df_taxa = adata.to_df("nona") + +# appending on the categorical data of interest +df_all = pd.concat([df_taxa, adata.obs[[batch_col, bio_col]]], axis=1).reset_index() + +df_all['bio'] = df_all['bio'].astype('category') +df_all['source'] = df_all['source'].astype('category') + +df_all.info() + +# %% [markdown] +# **`pd.DataFrame` Requirements for ABaCo:** +# +# The dataset contains the following making it compatible with the ABaCo framework: +# +# | id_col | batch_col | bio_col | count1 | count2 | ... | +# |--------|------------|----------|--------|--------|-----| +# | A | 24/07/2025 | RA | # | # | ... | +# | B | 15/06/2024 | RD | # | # | ... | +# | C | 24/07/2025 | RL | # | # | ... | +# +# - The data has categorical columns: +# 1. unique ids to identify the observations/samples e.g. sample id col +# 2. ids for the batch/factor groupings to be corrected by abaco. e.g. our phony bio data +# 3. biological/experimental factor variation for abaco to retain when correcting batch effect e.g., study id +# +# - And taxa counts to be trained on. +# +# We can use `abaco.plots.plotPCoA()` to visualize any batch and biological effects based on the given categories. + +# %% +from abaco.plots import plotPCoA +import plotly.io as pio +pio.renderers.default = "notebook" + +plotPCoA( + data=df_all, + sample_label=id_col, + batch_label=batch_col, + experiment_label=bio_col, +) + +# %% [markdown] +# - Batch effect (colours): +# - Different sequencing platforms which could result in a technical source of variation captured by the clustering of studies 6204 & 6205 (Illumina HiSeq 2500) vs. study 6231 (Illumina HiSeq 4000) along PCo1 +# +# - Biological effect (shapes): +# - The groups were randomly assigned so we expect no clustering by shape as supported by the pcoa above. This behaviour should not change with abaco reconstruction. +# ----- + +# %% [markdown] +# ## The goal +# +# Here the aim of **ABaCo** is to: +# 1) correct the batch effect (e.g., the points should no longer cluster by colour in the PCoA) while +# 2) maintaining biological variance (or lack thereof). +# +# Ideally, after using AbaCo to transform the data, the resulting PCoA coloured by batches will look like a colourful mixture of points. +# +# ----- +# +# ## Using `ABaCo` +# +# ### Setting up ABaCo +# +# We instantiate the `abaco.metaABaCo()` class and pass the required parameters shown in the cell below. +# +# Usually, setup of the parameters is required, which are explained in brief in the documentation e.g. `help(metaABaCo)` +# +# + +# %% +from abaco.ABaCo import metaABaCo +import torch + +device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + +# Instaantiate the ABaCo model +model = metaABaCo( + data=df_all, # Pre-processed dataframe + n_bios=df_all[bio_col].nunique(), # Number of biological groups in the data + bio_label=bio_col, # Column where biological groups are labeled in the dataframe + n_batches=df_all[batch_col].nunique(), # Number of batch groups in the data + batch_label=batch_col, # Column where batch groups are labeled in the dataframe + n_features=df_taxa.shape[1], # Number of features (taxonomic groups) + prior="MoG", # Prior distribution + device=device, # Device + d_z=16, # default dim of latent space + epochs = [3000, 1000, 3000], # num epochs for each training phase + disc_net=[256, 128, 64], # stronger discriminator + disc_act_fun=torch.nn.LeakyReLU(0.1), +) + +# %% tags=["hide-output"] +help(metaABaCo) + +# %% [markdown] +# ### Training the ABaCo model +# +# To train ABaCo on the prepared dataset, we then use method `abaco.metaABaCo.fit()`. + +# %% tags=["hide-output"] +# Train the model, +model.fit( + seed=42, + w_elbo_nll=10, # more emphasis on reconstruction + w_bio_penalty=0.0, # disable bio supervision + w_cluster_penalty=0.0, # light regularization or can be 0 + w_disc=1.0, # default, discriminator training strength +) + + +# %% [markdown] +# `abaco.metaABaCo` provides a method that visualizes the latent vectors via PCA for dimensionality reduction: `.plot_pca_posterior()` + +# %% +model.plot_pca_posterior() + +# %% [markdown] +# ### Reconstructing the dataset with ABaCo +# +# To reconstruct the dataset we use method `abaco.metaABaCo.correct()` and save it as a new layer `abaco` in the annotated dataset. + +# %% +# Reconstruct the dataset using the trained ABaCo model +corrected_dataset = model.correct(seed=42) +# save back to adata +adata.layers['abaco'] = corrected_dataset.set_index(id_col).drop(columns=[batch_col, bio_col]).values +adata.write_h5ad("data/mgnify_tomato.h5ad") + +# %% [markdown] +# ## Viewing the ABaCo reconstructed dataset +# +# Here we again take a look at the PCoA but first lets check out the mean differences between the original and batch corrected counts. + +# %% +# flatten +nona = adata.layers['nona'].flatten() +abaco = adata.layers['abaco'].flatten() +# calculate means and diffs +means = (nona + abaco) / 2 +diffs = nona - abaco +print(means.shape, diffs.shape) +# mean and limits of agreement +mean_diff = np.mean(diffs) +std_diff = np.std(diffs) +loa_upper = mean_diff + 1.96 * std_diff +loa_lower = mean_diff - 1.96 * std_diff + +# creating plot +fig = go.Figure() +fig.add_trace( + go.Scatter( + x=means, y=diffs, + mode='markers', + marker=dict(color='blue', opacity=0.1) + ) +) +fig.add_hline(y=mean_diff,line_color="black", annotation_text="Mean Diff") +fig.add_hline(y=loa_upper, line_dash="dot", line_color="black") +fig.add_hline(y=loa_lower, line_dash="dot", line_color="black") +# titles +fig.update_layout( + template="plotly_white", + title="Bland-Altman Plot: Original vs. ABaCo Corrected", + xaxis_title="Mean of Original and ABaCo Corrected", + yaxis_title="Difference (original - abaco)", +) +fig.show() + +# %% [markdown] +# No clear tendency for abaco correction to be lower or higher than original counts. + +# %% +# Plot the PCoA of the reconstructed dataset +plotPCoA( + data = corrected_dataset, + sample_label=id_col, + batch_label=batch_col, + experiment_label=bio_col +) + +# %% [markdown] +# In the PCoA we see that there is more overlap of the batches based on the colours mixing. + +# %% [markdown] +# ## Conclusion +# +# The goal was to: +# +# ✅ correct the batch effect (reduce clustering of points by colour in the PCoA and pairplot coloured by batches) +# +# A brief visual inspection of the PCoA suggests that ABaCo reduced the batch effect associated with analyzing the samples from different instruments, while still maintaining the original biological variance. +# +# +# --- + +# %% [markdown] +# diff --git a/docs/tutorial/tutorial-anaerobic-digestion.py b/docs/tutorial/tutorial-anaerobic-digestion.py index e9b2130..270407b 100644 --- a/docs/tutorial/tutorial-anaerobic-digestion.py +++ b/docs/tutorial/tutorial-anaerobic-digestion.py @@ -5,7 +5,7 @@ # extension: .py # format_name: percent # format_version: '1.3' -# jupytext_version: 1.17.2 +# jupytext_version: 1.18.1 # kernelspec: # display_name: .venv # language: python diff --git a/requirements.txt b/requirements.txt index 2d86214..63bdb1e 100644 --- a/requirements.txt +++ b/requirements.txt @@ -55,6 +55,7 @@ python-dateutil pytz PyYAML requests +scanpy scikit-bio scikit-learn scipy diff --git a/src/abaco/ABaCo.py b/src/abaco/ABaCo.py index dd93495..9379b6f 100644 --- a/src/abaco/ABaCo.py +++ b/src/abaco/ABaCo.py @@ -3567,7 +3567,12 @@ def cluster_loss(self): kl_matrix[i, j] = kl # Take the minimum KL divergence between any two components - min_kl = kl_matrix[kl_matrix > 0].min() + positive_kl = kl_matrix[kl_matrix > 0] + if positive_kl.numel() == 0: + # Handle the case where there are no positive KL values + min_kl = 0.0 # or float('inf'), or another default + else: + min_kl = positive_kl.min() # Loss is inverse of min KL (maximize separation) return 1.0 / (min_kl + 1e-8) @@ -3702,7 +3707,12 @@ def cluster_loss(self): kl_matrix[i, j] = kl # Take the minimum KL divergence between any two components - min_kl = kl_matrix[kl_matrix > 0].min() + positive_kl = kl_matrix[kl_matrix > 0] + if positive_kl.numel() == 0: + # Handle the case where there are no positive KL values + min_kl = 0.0 # or float('inf'), or another default + else: + min_kl = positive_kl.min() # Loss is inverse of min KL (maximize separation) return 1.0 / (min_kl + 1e-8) diff --git a/src/abaco/utils.py b/src/abaco/utils.py index 3c132f0..b5c2b54 100644 --- a/src/abaco/utils.py +++ b/src/abaco/utils.py @@ -5,6 +5,7 @@ import sys from datetime import datetime from urllib.parse import urlparse, urlunsplit +import pandas as pd # import yaml @@ -456,3 +457,50 @@ def get_logger(): # assert isinstance(contents, dict), "content not returned as a dict" # return contents + + +def df_joiner( + df_dict: dict[pd.DataFrame], + on: str, + how: str = "outer", +) -> pd.DataFrame: + """ + Join multiple dataframes on a common column. + + Parameters + ---------- + df_dict : dict of pandas.DataFrame + Dictionary of dataframes to join. + on : str, optional + Column to join on. Defaults to "taxa". + how : str, optional + Type of join. Defaults to "outer". + + Returns + ------- + pandas.DataFrame + Joined dataframe. + """ + + ## PRECONDITION CHECKS + if not isinstance(df_dict, dict): + raise TypeError(f"df_dict must be a dict: {type(df_dict)}") + if not isinstance(on, str): + raise TypeError(f"on must be a str: {type(on)}") + for key, df in df_dict.items(): + if not isinstance(df, pd.DataFrame): + raise TypeError(f"df_dict values must be pd.DataFrame: {type(df)}") + if (on not in df.columns) and (on not in df.index.names): + raise ValueError(f"Column '{on}' not found in dataframe with key '{key}'") + if how not in ["left", "right", "outer", "inner"]: + raise ValueError(f"how must be one of 'left', 'right', 'outer', 'inner': {how}") + + ## MAIN FUNCTION + # dfs into a list + df_list = list(df_dict.values()) + # init the merged df with the first one + df_merged = df_list[0] + # for all others, merge iteratively + for df in df_list[1:]: + df_merged = pd.merge(df_merged, df, on=on, how=how) + return df_merged