# MMDRP: Drug Response Prediction and Biomarker Discovery Using Multi-Modal Deep Learning

![Figure_2](https://github.com/LincolnSteinLab/MMDRP/assets/21085967/7d28b23e-e06d-4177-ad20-13ca37c40c49)

MMDRP is now published in [Bioinformatics Advances](https://doi.org/10.1093/bioadv/vbae010)! (Open Access)

This repository contains preprocessing, training and evaluation code for MMDRP models.

## Preprocessing

Training data was obtained from the following:

- CTRPv2 was obtained and processed using the PharmacoGx BioConductor package (https://bioconductor.org/packages/release/bioc/html/PharmacoGx.html)

  * Please refer to the `R/01_Dose-Response_Data_Preparation.R` file for details.

- DepMap Portal (https://depmap.org/portal/) for cell line profiling data.

  * 20Q2 for Protein Quantification data (lastest) and 21Q2 for mutational, gene expression, CNV, miRNA, metabolomics, histone, and RPPA data.

  * Please refer to the `R/02_Omic_Data_Preparation.R` file for details.

## Training

Training was done in Python using the Pytorch framework. `.py` files are available in the `src` folder.  

`drp_full_model.py`is the main file used for training which can be run as a commandline program. Please refer to this file for the list of input arguments and their defaults.  

## Evaluation

Evaluation was performed using multiple cross-validation schemes. The predictions from the validation sets were then aggregated for each model, and further analyzed and compared in the `05_All_Comparison_Plots.R` file.