GADRP

Source code and data for "GADRP: graph convolutional networks and autoencoders for cancer drug response prediction"

Data

• 269-dim-physicochemical.csv - Physicochemical properties of drugs
• 881-dim-fingerprint.csv - Molecular fingerprint of drugs
• miRNA_470cell_734dim.csv - MicroRNA expression data of cell lines
• CpG_407cell_69641dim.csv - DNA methylation data of cell lines
• RNAseq_462cell_48392dim.csv - Gene expression data of cell lines
• copynumber_461cell_23316dim.csv - DNA copy number data of cell lines
• drug_cell_response.csv - response data between drugs and cell lines
• cell_name.csv - Names of 388 cell lines with four cell line characteristics

Source codes

• drug.py: generate drug similarity matrix according to physicochemical properties of drugs
• cell.py: generate cell line similarity matrix according to microRNA expression and DNA methylation of cell lines
• drug_cell.py: generate drug cell line pairs similarity matrix
• cell_ae.py: learn low_dimensional representations from high-dimensional cell line features
• train.py: train the model and make predictions
• GADRP.py: details of GADRP model

Requirements

• Python == 3.7.10
• PyTorch == 1.9.0
• sklearn == 0.24.2
• Numpy == 1.19.2
• Pandas == 1.3.4

Operation steps

1. Install dependencies, including torch1.9, sklearn, numpy and pandas
2. run drug.py and cell.py to generate drug and cell line similarity matrices
3. run drug_cell.py to generate drug cell line pair similarity matrix
4. run cell_ae.py to generate low-dimensional representations of cell line omics characteristics
5. run python train.py for training and prediction

Installation

git clone https://github.com/flora619/GADRP