Skip to content

albertcyhe/nets-function

BranchesTags

Repository files navigation

Evidence Chain for NET Marker–Function Decoupling in Brain Metastasis (with reproducible records)

This README consolidates and reconstructs prior work across RNA and Proteomics/Metabolomics layers, following the flow “rationale → execution → results”, and preserves failures/adjustments for transparency.

Acknowledgments

  • Inspired by Openmind Club.
  • Special thanks to Zhiping Yang for insightful discussions and encouragement.

Quick Start

  • One‑click data retrieval: bash scripts/sh/download_all_data.sh (see DATA_SOURCES.md for coverage and notes)
  • One‑click environments: bash scripts/sh/setup_proteomics_env.sh and bash scripts/sh/setup_scrna_env.sh (see ENVIRONMENT.md)
  • Activate proteomics env: source env/activate_proteomics.sh
  1. Motivation and Hypotheses (markers ≠ function; decoupling?)
  • Background: Prior work suggests CTSC‑induced inflammation → NETosis → NET release → TSP1 (THBS1) cleavage and pro‑metastatic effects; many studies strongly bind NET “markers (NET‑M)” to “function (NET‑F)”.
  • Core question: markers are not function — can they decouple? In acute inflammation literature, “markers < function” is often observed. If in metastasis prevention markers and function can be separated, clinical implications and treatment stratification are very different.
  • Quantitative definition: ΔFM = z(F) − z(M)
    • NET‑F (functional axis): ELANE/PRTN3/CTSG (avoid upstream ROS generators).
    • NET‑M (marker axis): anchor at PADI4, expand by co‑expression in neutrophil‑enriched samples (v2; designed to be anti‑correlated with F).
  • Hypothesis: decoupling is brain‑microenvironment‑specific, manifesting as “markers > function” (ΔFM decreases), coupled to THBS1/ECM/angiogenesis external‑growth axes.

Statistical standards and thresholds

  • Tests: paired Wilcoxon (report Cliff’s δ), unpaired Mann‑Whitney/linear models; correlations via Spearman + BH FDR (alpha = 0.05).
  • Partial correlations/mixed models: control Neutrophil score and TumorPurity; stratify or include (1 | cohort/dataset).
  • Reporting: two‑sided tests, FDR < 0.05; report effect sizes with 95% CI; random‑effects meta‑analysis (DL), I² < 50% considered consistent.

Summary and next steps

  • Result highlights: propose ΔFM as a quantitative indicator of “marker–function imbalance”, define NET‑F/NET‑M and statistical standards, creating a unified framework for RNA/protein validation.
  • Clinical significance: if “markers > function” brain‑specific decoupling holds, prevention and follow‑up stratification are affected (functional anchors require separate monitoring).
  • Next steps: validate ΔFM stability in more independent cohorts, refine NET‑M v2 and negative controls, and finalize a locked resource list.
  1. Transcriptome evidence: lung coupling, brain decoupling (markers > function)

  • Data and design (bulk)

    • Paired metastasis: GSE184869 (RNA‑seq; log2 TMM CPM provided), GSE125989 (Affy GPL571).
    • Metastasis stratification: GSE43837 (brain), GSE14017/14018 (non‑brain).
    • Covariates: Neutrophils (MCPcounter), TumorPurity (tidyestimate/ESTIMATEScore → TumorPurity).
    • Scoring: singscore and ssGSEA (GSVA ≥ 2.2 param API; de‑duplicate and aggregate rownames). Compute ΔFM after within‑sample z‑scaling.

  • Main results (bulk)

    • ΔFM paired difference (met − primary)
      • GSE184869 (RNA‑seq): singscore Wilcoxon V=14, p=2.10e−4, FDR=3.22e−4, Cliff’s δ≈−0.70, n=20; ssGSEA consistent in direction.
        • Figure: results/figures/Figure2_GSE184869_Paired_DFM.pdf
        • Table: results/tables/GSE184869_paired_tests.tsv
      • GSE125989 (Affy): V=91, p=0.252, FDR=0.298, Cliff’s δ≈0.25, n=16 (ns).
        • Table: results/tables/GSE125989_paired_tests.tsv
    • ΔFM and THBS1 partial correlation (control Neutrophil + TumorPurity)
      • GSE184869: singscore ρ=−0.3004, p=0.00415, FDR≈0.00831, n=90; ssGSEA ρ=−0.2732, p=0.00936, FDR≈0.00936.
        • Figure: results/figures/Figure3_GSE184869_THBS1_Residuals.pdf
        • Table: results/tables/GSE184869_THBS1_residuals.tsv (partial‑corr backup: results/tables/GSE184869_thbs1_partial_cor.tsv)
      • GSE125989: ρ=0.2269, p=0.2109, FDR≈0.3149 (ns).
    • F/M correlation across cohorts (5 cohorts × 2 methods)
      • GSE184869: singscore ρ=−0.2081, p=0.0491, FDR=0.098; ssGSEA ρ=−0.1216, p=0.2537.
      • GSE125989: ρ=0.4941, p=0.00449, FDR=0.01495; ssGSEA ρ=0.4025, p=0.02314.
      • GSE43837: ρ≈0.07–0.11; GSE14017: ρ≈0.23–0.26; GSE14018: ρ≈0.65–0.67 (p ≪ 0.001).
        • Figure: results/figures/Figure4_FM_Corr_Overview.pdf
        • Table: results/tables/fm_correlation_overview.tsv

  • Single‑cell (mechanistic carrier) GSE186344

    • After strict gating (high NEUT≥0.70 and low TNK/Mono≤0.60), no stable neutrophil cluster was found (or counts ~0), thus ΔFM.cell within neutrophils and a “low‑ΔFM” subpopulation could not be established.
    • Interpretation: consistent with “neutrophils have lysed to NETs and functional axis is suppressed”; scRNA captures intracellular transcription and is less suited for extracellular/degradation footprints.
      • Figure: results/figures/Figure5_scRNA_LowDFM_Fraction.pdf (placeholder interpretation noted in logs)
      • Record: results/figures/Figure5_scRNA_THBS1_Assoc.txt (non‑estimable correlation due to size/constant issues)
      • Tables: results/tables/gse186344_neutrophil_lowDFM_fraction.tsv; results/tables/gse186344_sample_subpop_fraction.tsv; results/tables/gse186344_subpop_thbs1_assoc.tsv

  • Execution scripts (transcriptomics)

  • R: scripts/r/02_covariates_estimate.R, scripts/r/03_scores_deltafm.R, scripts/r/04_bulk_stats_mvp.R, scripts/r/05_pathway_scores.R, scripts/r/06_fm_correlation_overview.R

  • Figures (re‑draw): Rscript scripts/r/13_figures_a3.R

  • Modules/resources: Rscript scripts/r/export_selected_sets.R

  • Outreach: python scripts/py/met500_fetch.py, python scripts/py/site_triage_met500.py (MET500 resources)

  1. Proteomics Evidence (A3): Methods and Results

  • Datasets

    • PRIDE PXD011796 (NET reference; label‑free), PXD005719 (brain‑metastatic membrane proteome), PXD046330 (HER2+ conditioned media secretome), PXD051579 (HER2+ brain‑metastasis BBB; paused for cross‑omics), plus MassIVE MSV000089062 and PRIDE PXD018301/PXD032767 for liquid prototypes.

  • Processing and scoring (high level)

    • Convert RAW→mzML; search with MSFragger; infer proteins with Philosopher; normalize and compute per‑sample module scores.
    • Derived endpoints: Proteo‑ΔFM (protein ΔFM analogue), NE/PR3 substrate footprints (semi/ non‑tryptic N‑termini proxy), THBS1 cleavage index, Serpin scores (CLR), MNAR detection indicators.

  • A3 model executions — first pass (strict evidence)

    • ISI (Inhibitor–Protease Stoichiometry): scripts/py/isi_model.py → results/tables/{isi_per_sample.tsv, isi_models.tsv}.
      • Result: In current cohorts (PXD046330, PXD005719, PXD051579), NET‑F targets (ELANE/PRTN3/CTSG) are not detected (target_sum=0), so ISI is undefined by design, consistent with “F suppressed/not measurable while M persists”.
    • MNAR (Missingness‑as‑signal): scripts/py/mnar_detection_model.py → results/tables/{mnar_detection_table.tsv, mnar_logit_results.tsv}.
      • Result: detect_F=0 across samples; detect_M=1 across samples → non‑estimable logits but a definitive qualitative pattern: “F fully missing, M universally present”.
    • Footprint indices (substrate‑centric NE/PR3 proxy): scripts/py/footprint_indices.py → results/tables/footprints.tsv.
      • Result: indices are consistently negative; in PXD005719, brain‑variant < parental (≈−0.786 vs −0.633); PXD046330/PXD051579 footprints are negative, consistent with low NE/PR3 activity.

  • Stratified associations (evidence chain)

    • Footprints vs Serpin (expected negative)
      • Combined: ρ ≈ −0.563, p ≈ 0.00121, q ≈ 0.00182; Non‑brain: ρ ≈ −0.560, p ≈ 0.00160, q ≈ 0.00130; Brain: under‑powered.
    • THBS1 abundance vs Serpin (expected positive)
      • Combined: ρ ≈ +0.583, p ≈ 0.00073, q ≈ 0.00218; Non‑brain: ρ ≈ +0.611, p ≈ 0.00043, q ≈ 0.00130.
    • Proteo‑ΔFM vs Serpin (expected negative)
      • Combined/Non‑brain: weak/non‑significant (ρ ≈ +0.225, p ≈ 0.29), consistent with M‑driven ΔFM under F missingness.
    • Outputs: results/tables/assoc_summary.tsv (per‑stratum summaries).

  • Mixed‑effects models

    • scripts/py/mixed_effects_a3.py → results/tables/mixed_effects_a3.tsv
      • Formula: endpoint ~ Serpin_score_core * organ + log_total_PSMs + (1 | dataset)
      • THBS1 abundance fit: β_Serpin ≈ +0.028 (ns; limited organ contrast); footprint model singular (brain underpowered); messages annotated.
    • scripts/py/mixed_effects_growth.py (non‑brain focus) → results/tables/mixed_effects_growth.tsv
      • Footprints ~ Serpin_score_core: β ≈ +0.0027 (p ≈ 4e−20) reflecting monotone footprint drop as Serpin increases (index negative, so small positive slope = stronger suppression).
      • THBS1 cleavage ~ Serpin: β ≈ −0.011 (p ≈ 0.41); THBS1 log abundance ~ Serpin: β ≈ +0.0125 (p ≈ 0.30).
      • Warnings (brain underpowered, singular) recorded in results/tables/mixed_effects_messages.txt.

  • SEM prototype (exploratory)

    • scripts/py/prepare_sem_data.py → results/tables/sem_input.tsv
    • scripts/r/14_sem_model.R (lavaan) → results/models/sem_results.json, results/figures/A3_sem_path.svg
      • Model: Serpin_score_core → F_latent → {footprint, THBS1_cleave, Proteo‑ΔFM}; standardized paths saved; warnings indicate limited identifiability (n≈24).

  • Multi‑omics alignment (status)

    • results/tables/multiomics_alignment.tsv and results/tables/multiomics_pairwise_deltas.tsv align PXD046330 ↔ GSE96860 (SKBR3) and PXD005719 ↔ GSE12237 (MDA‑MB‑231 parental vs brain variants). Spearman ΔFM correlations underpowered (n<3 per cohort); NET‑M/THBS1 deltas remain interpretable; PXD051579 cross‑omics paused until a matched transcriptomic dataset exists.
  1. Site Triage (Liquid Prototypes)
  • CSF (MSV000089062, n≈72; BrainMet≈17)
    • Features: data/processed/proteomics/MSV000089062/csf_patient_features.tsv; outputs include metrics/ROC/PR/calibration/DCA/coefficients/scaler/spearman/predictions (see “Figures and tables”).
  • Plasma‑EV (PXD018301, n≈512; BrainMet≈12)
    • Multi‑level Excel (Human512Reports) normalized by Area; columns renamed to sample names; outputs include full metrics and robust logit.
  • Plasma (PXD032767; MaxQuant)
    • Non‑brain RAW not public; use Supplementary Table S1 to map groups; construct run_manifest.tsv; compute footprints/THBS1 with MaxQuant exports.
    • Clinical: data/interim/proteomics/PXD032767/metadata/clinical_metadata.tsv
    • Manifest: data/interim/proteomics/PXD032767/metadata/run_manifest.tsv
    • Intermediates: data/interim/proteomics/PXD032767/metadata/{footprint_maxquant.tsv, thbs1_cleavage.tsv}
  • Fusion (PXD032767 × PXD018301)
    • Add dataset indicators; output robust logit and (1|dataset) mixed effects alongside DCA.
  • Modelling notes
    • Robust logit: implement HC3 manually (leverage h, residuals) when library variant lacks HC3 for logistic.
    • Mixed effects: BinomialBayesMixedGLM; report fe_mean/fe_sd; directions match robust logit.
    • ROC/effects for CSF/Plasma deliver AUC ~0.60–0.79 usable range.
  1. Metabolomic Extensions (Paused)
  • Current Workbench studies (ST000745, ST001104, ST002921) expose mwTab metadata and/or instrument RAW only; named metabolite matrices are not available; redox ratios (e.g., GSH/GSSG) not computable yet.
  • Scripts archived pending restoration once named metabolite tables are available.
  1. Cross‑Layer Integration & Modelling (overview)
  • Sample alignment: results/tables/multilayer_sample_map.tsv linking cohort/organ/patient/sample types and available layers.
  • Derived metrics table: results/tables/multilayer_features.tsv combining RNA ΔFM, Proteo‑ΔFM, Serpin score, THBS1 cleavage index, ROS metrics, MPO traces (where available).
  • Mixed‑effects model (R, lme4 + lmerTest): Functional_proxy ~ DeltaFM * Organ + Proteo_DeltaFM + Serpin_score + ROS_ratio + (1 | Cohort); evaluate organ slopes via emmeans.
  • Visual storytelling: “double sandwich” schematic results/figures/F_double_sandwich.pdf.
  1. Failures and adjustments (transparent records)
  • Transcriptome: scRNA THBS1 association not estimable due to size/constant issues; strict gating did not yield a stable neutrophil cluster (or near‑zero), thus “low‑ΔFM neutrophil fraction” cannot be robustly estimated — consistent with “neutrophils already lysed to NETs; functional axis suppressed”.
  • Protein/peptide: NET‑F targets are sparse; ISI and robust logit non‑estimable in some cohorts; pivot to “missingness‑as‑signal + footprints/THBS1 cleavage” as twin anchors for function.
  • Mixed effects: statsmodels==0.14.5 still reports singular warnings (expected with low n); coefficient directions stable; notes in results/tables/mixed_effects_messages.txt.
  • SEM: n≈24, limited identifiability (non‑invertible information matrix, negative variances); treat as conceptual.
  • Metabolomics: Workbench (ST000745/ST001104/ST002921) provide mwTab meta + RAW only; named metabolite matrices unavailable; redox ratios (GSH/GSSG) not computable yet; scripts archived pending restoration.
  1. Data, directories, and environments (conventions)

  • Directory layout

    • Raw: data/raw/<dataset>/ (keep checksums.sha256)
    • Intermediates: data/interim/<modality>/<dataset>/ (mzML, search workspaces, temporary TSVs)
    • Processed: data/processed/<dataset>/ (gene/protein matrices, cleavage indices, covariates)
    • Resources: resources/modules/, resources/manifests/, resources/annotation/ (includes gencode.v43.basic.annotation.gtf.gz), resources/msfragger/
    • Results: results/tables/, results/figures/

  • Environments and tools

    • Proteomics env: env/proteomics.yml; activate with:
      • export MAMBA_ROOT_PREFIX="$PWD/env/.mamba"
      • env/bin/micromamba env create -y -f env/proteomics.yml
      • eval "$(env/bin/micromamba shell hook -s bash)"
      • micromamba activate proteomics
      • export PATH="$PWD/env/bin:$PATH"; philosopher version (sanity check)
    • RNA bigWig: includes pybigwig and gffread; annotation at resources/annotation/gencode/gencode.v43.basic.annotation.gtf.gz
    • Optional R user library: R_LIBS_USER=env/.R (msqrob2/lmerTest/plot, etc.)
    • Long jobs: tmux/screen; log to logs/ (e.g., logs/proteomics/pipeline_<timestamp>.log)
    • Scripts: scripts/sh/setup_proteomics_env.sh (one‑click proteomics deps); RAW→mzML: scripts/sh/run_trfp_parallel.sh; smoke test: scripts/sh/msfragger_smoke.sh; GEO supplementary: scripts/sh/fetch_gse186344_suppl.sh.

  • Data retrieval (examples)

    • PRIDE: bash scripts/sh/fetch_pride_dataset.sh PXD011796 --categories RAW,RESULT --dest data/raw/PXD011796
    • PXD005719: scripts/sh/fetch_pride_dataset.sh PXD005719 --categories RAW,RESULT --dest data/raw/PXD005719
    • PXD046330: scripts/sh/fetch_pride_dataset.sh PXD046330 --categories RAW,RESULT --dest data/raw/PXD046330
    • PXD051579 (paused): scripts/sh/fetch_pride_dataset.sh PXD051579 --categories RAW,RESULT --dest data/raw/PXD051579
    • GSE96860 bigWig: python scripts/py/process_gse_bigwig.py --bigwig-dir data/raw/transcriptomics/GSE96860 --gtf resources/annotation/gencode/gencode.v43.basic.annotation.gtf.gz --dataset GSE96860 --output-dir data/processed/GSE96860
    • GSE12237 array: python scripts/py/parse_series_matrix.py --series data/raw/transcriptomics/GSE12237/GSE12237_series_matrix.txt.gz --gpl data/raw/transcriptomics/GSE12237/GPL96.annot.gz --dataset GSE12237 --outdir data/processed/GSE12237
    • CSF proteomes (various): scripts/sh/fetch_pride_dataset.sh <CSF_ACCESSION> --categories RESULT --match 'CSF' --dest data/raw/<CSF_ACCESSION>
    • CSF metabolomics: aria2c -c -x16 -s16 --max-tries=0 --retry-wait=30 -i resources/manifests/metabolomics_urls.txt -d data/raw/metabolomics
  1. Reproducible script entry points and suggested order

  • Bulk (MVP)

    • Rscript scripts/r/02_covariates_estimate.R --dataset GSE184869 --config resources/config.yml
    • Rscript scripts/r/03_scores_deltafm.R --dataset GSE184869 --config resources/config.yml
    • Rscript scripts/r/04_bulk_stats_mvp.R --dataset GSE184869 --config resources/config.yml
    • Rscript scripts/r/05_pathway_scores.R --dataset GSE184869 --config resources/config.yml
    • Rscript scripts/r/06_fm_correlation_overview.R --config resources/config.yml

  • scRNA (GSE186344)

    • python scripts/py/sc_preprocess_gse186344.py --raw GSE186344_RAW.tar --out data/processed/scRNA/gse186344.h5ad
    • python scripts/py/sc_neutrophil_subsets.py --h5ad data/processed/scRNA/gse186344.h5ad --net_f resources/modules/net_f_v1.tsv --net_m resources/modules/net_m_v2.tsv --outdir results/tables
    • python scripts/py/sc_tumor_thbs1_assoc.py --h5ad data/processed/scRNA/gse186344.h5ad --frac results/tables/gse186344_neutrophil_lowDFM_fraction.tsv --out results/tables/gse186344_subpop_thbs1_assoc.tsv

  • Proteins/peptides (A3)

    • python scripts/py/footprint_indices.py → results/tables/footprints.tsv
    • python scripts/py/thbs1_cleavage_index.py → results/tables/thbs1_cleave_idx.tsv
    • python scripts/py/serpin_scores.py → results/tables/serpin_scores.tsv
    • python scripts/py/proteomics_deltafm_score.py → data/processed/proteomics/<dataset>/proteo_deltafm.tsv
    • python scripts/py/isi_model.py → results/tables/isi_per_sample.tsv, isi_models.tsv
    • python scripts/py/mnar_detection_model.py → results/tables/mnar_detection_table.tsv, mnar_logit_results.tsv
  1. Figures and tables
  1. Next steps
  • ISI (core/extended) + organ interaction; export forest plots.
  • Logit/zero‑inflation: F detection ~ Serpin (M as negative control); meta‑analysis across cohorts.
  • Substrate footprints and THBS1 cleavage: robust estimates and visualization against Serpin/Proteo‑ΔFM.
  • Liquid prototype page: integrate CSF/Plasma ROC and effect sizes; update brain vs non‑brain mechanism schematic.

About

Analysis of comparison of M&F of NETs

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages