a b/README.md 

  1
  
# MMDRP: Drug Response Prediction and Biomarker Discovery Using Multi-Modal Deep Learning





  
  

  2
  
![Figure_2](https://github.com/LincolnSteinLab/MMDRP/assets/21085967/7d28b23e-e06d-4177-ad20-13ca37c40c49)





  
  

  3
  





  
  

  4
  





  
  

  5
  
MMDRP is now published in [Bioinformatics Advances](https://doi.org/10.1093/bioadv/vbae010)! (Open Access)





  
  

  6
  





  
  

  7
  
This repository contains preprocessing, training and evaluation code for MMDRP models.





  
  

  8
  





  
  

  9
  
## Preprocessing





  
  

  10
  
Training data was obtained from the following:





  
  

  11
  
- CTRPv2 was obtained and processed using the PharmacoGx BioConductor package (https://bioconductor.org/packages/release/bioc/html/PharmacoGx.html)





  
  

  12
  
  * Please refer to the `R/01_Dose-Response_Data_Preparation.R` file for details.





  
  

  13
  
- DepMap Portal (https://depmap.org/portal/) for cell line profiling data.





  
  

  14
  
  * 20Q2 for Protein Quantification data (lastest) and 21Q2 for mutational, gene expression, CNV, miRNA, metabolomics, histone, and RPPA data.





  
  

  15
  
  * Please refer to the `R/02_Omic_Data_Preparation.R` file for details.





  
  

  16
  
## Training





  
  

  17
  
Training was done in Python using the Pytorch framework. `.py` files are available in the `src` folder.  





  
  

  18
  
`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.  





  
  

  19
  





  
  

  20
  
## Evaluation





  
  

  21
  
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.