Cross-Study Benchmark Dataset for Monotherapy Drug Response Prediction

This benchmark dataset was created and used to train and evaluate models presented in the paper: A. Partin, P. Vasanthakumari et al., "Benchmarking community drug response prediction models: datasets, models, tools, and metrics for cross-dataset generalization analysis." This dataset includes four main components: cell line drug response data, cell line multi-omics data, drug feature data, and predefined data partitions for modeling. Data response data were curated from five pharmacogenomic studies (CCLE, CTRPv2, GDSC1, GDSC2, GCSI), and processed using a unified pipeline for response fitting, omics harmonization, and drug representation. Multi-dose viability data were extracted, and a unified dose response fitting pipeline was used to calculate multiple dose-independent response metrics, such as the area under the dose response curve (AUC) and the half-maximal inhibitory concentration (IC50). The multi-omics data of cell lines were extracted from the the Dependency Map (DepMap) portal of CCLE, including gene expressions, DNA mutations, DNA methylation, gene copy numbers, protein expressions measured by reverse phase protein array (RPPA), and miRNA expressions. Data preprocessing was performed, such as discretizing gene copy numbers and mapping between different gene identifier systems. Drug information was retrieved from PubChem. Based on the drug SMILES strings, we calculated their molecular fingerprints and descriptors using the Mordred and RDKit Python packages. Data partition files were generated using the IMPROVE benchmark data preparation pipeline. They indicate, for each modeling analysis run, which samples should be included in the training, validation, and testing sets, for building and evaluating the drug response prediction (DRP) models. More detailed information about the dataset and its construction can be found at https://jdacs4c-improve.github.io/docs/content/app_drp_benchmark.html