Motivation

In many patients, a tumor’s innate or acquired resistance to a given therapy will render the treatment ineffective. To increase therapeutic options and to overcome drug resistance, cancer researchers have been actively investigating drug combinations.

Build environment (dependencies)

conda env create -f environment.yml

Check model performance with differnt cell line features (gene dependencies, gene expression and netexpress scores)

gene dependencies

cp setting_gene_dependencies.py setting.py

gene expression

cp setting_gene_expression.py setting.py

netexpress

cp setting_net.py setting.py

Run

python ./attention_main.py

check results

check the logfile in the newest _run_***** folder

Dataset

Download processed dataset

1. Install zenodo-get

pip install zenodo-get

2. Download data to src/synergy_score file

zenodo_get 10.5281/zenodo.4789936

Drug combination Synergy scores

An Unbiased Oncology Compound Screen to Identify Novel Combination Strategies. (O'Neil J et. al)

Unnamed: 0,drug_a_name,drug_b_name,cell_line,synergy
5-FU_ABT-888_A2058,5-FU,ABT-888,A2058,7.6935301658
5-FU_ABT-888_A2780,5-FU,ABT-888,A2780,7.7780530601
5-FU_ABT-888_A375,5-FU,ABT-888,A375,-1.1985054379
5-FU_ABT-888_A427,5-FU,ABT-888,A427,2.5956844375
5-FU_ABT-888_CAOV3,5-FU,ABT-888,CAOV3,-5.1399712212
5-FU_ABT-888_DLD1,5-FU,ABT-888,DLD1,1.9351271188

Cell line gene-level dependencies score

index,genes,CAL148,HT1197,A2780,MCF7,HCC1395, ..., ENTREZID
CDH2 (1000),CDH2,0.1709,0.2125,0.1651,-0.0103,-0.2448, ..., 1000
AKT3 (10000),AKT3,-0.0400,-0.0137,0.2180,-0.0194],-0.0503644515648, ..., 10000

Drug features

Drug target profile

Gene networks

Entrezid Entrezid posterior prob.
5988    53905   0.137373
5988    286234  0.116511
5988    277     0.104127
5988    387856  0.114427
5988    90317   0.115751
5988    100287366       0.100036
5988    100287362       0.105938