Integrative pharmacogenomics to infer large-scale drug taxonomy

Please cite: Integrative Cancer Pharmacogenomics to Infer Large-Scale Drug Taxonomy. El-Hachem N, Gendoo DMA, Ghoraie LS, Safikhani Z, Smirnov P, Chung C, Deng K, Fang A, Birkwood E, Ho C, Isserlin R, Bader GD, Goldenberg A, Haibe-Kains B. Cancer Res. 2017 Jun 1;77(11):3057-3069. doi: 10.1158/0008-5472.CAN-17-0096. Epub 2017 Mar 17. PMID: 28314784

This project assess the utility of data integration from multiple drug information layers.

Hypothesis: Better characterization of compound mechanism of action (MoA) from integrative pharmacogenomics.

Impact: Identifying mechanism for compounds with unknown MoA in early stages of drug development without the need of sophisticated info (side effects or pharmacological indications...)

Based on the method described in

Similarity network fusion for aggregating data types on a genomic scale. Wang B, Mezlini AM, Demir F, Fiume M, Tu Z, Brudno M, Haibe-Kains B, Goldenberg A. Nat Methods. 2014 Mar;11(3):333-7. doi: 10.1038/nmeth.2810. Epub 2014 Jan 26. PMID: 24464287

Reproducibility of the Analysis

The following steps will reproduce the output files (figure, tables...) mentioned in the main manuscript. The script will be using data files such as:

Files to run the scripts

• Drug-target (benchmark) from CHEMBL and CTRP
• Sensitivity measurements (drugs x cell lines) from CTRPv2 and NCI-60
• Transcriptomic data from the LINCS database http://lincs.hms.harvard.edu/ created using PharmacoGx package https://cran.r-project.org/web/packages/PharmacoGx/index.html

Run the R scripts

• main-ctrpv-lincs.R and
• main-nci60-lincs.R

(this will execute the following R codes):

# process the raw data, find common drugs...
preprocessInput.R 	

# remove problematic drugs and missing data
sensitivityData.R 	

# get the structural fingerprints (extended-connectivity descriptors from RCDK)
structureData.R 

# remove problematic drugs and missing data and find drug names from LINCS metadata
perturbationData.R 	

# Get the similarity matrix from structure (tanimoto metric)
constStructureLayer.R 	

# Get the similarity matrix from sensitivity (pearson metric), rescale to 0-1 with function in SNF
constSensitivityLayer.R	

# Get the similarity matrix from perturbation (pearson metric), rescale to 0-1 with SNF affinitymatrix
constPerturbationLayer.R

# Integrate all 3 layers using SNFtools
integrateStrctSensPert.R	

# ATC code gold standard benchmark from CHEMBL
ATCbench.R	

# Drug-target benchmarks from CHEMBL and CTRPv2
drugTargetBench.R	

# get the pairs of drugs sharing the same target (0 or 1)
generateDrugPairs.R	

# Compute concordance index between benchmark data and similarity network data
compConcordIndx.R	

# Get the p-value between concordance indices
predPerf.R	

# Generate ROC plots and AUC values
generateRocPlot.R	

# Community clustering from apcluster package in R
communityGen.R

# Network based on exemplar drugs from apcluster
main-network-generation.R	

Set up the software environment

A Docker Terminal Environment has been setup with all dependencies installed. Access it by installing Docker Engine and running the following lines of code:

#clone image
docker pull gosuzombie/dnf

#run image in interactive command line 
docker run -it gosuzombie/dnf

Alternatively, one can setup the Docker environment by using the provided Dockerfile. Instructions can be found here