DeepDRK

DeepDRK predicts anticancer drug responses and prioritizes drug repurposing by integrating genomics, transcriptomics, epigenomics, chemical properties, and drug-target interaction data using kernel-based integration and deep neural networks.

Key Features:

• Kernel-Based Multi-Omics Integration: Uses kernel-based similarity matrices to integrate genomics, transcriptomics, epigenomics, chemical properties, and known drug-target interactions.
• Deep Learning Framework: Employs deep neural networks trained on over 20,000 pan-cancer cell line-anticancer drug pairs to transfer information across drugs and cancer types.
• Performance and Robustness: Demonstrated superior accuracy and robustness on benchmark cancer cell line datasets and on newly established patient-derived cancer cell lines (AUC 0.84, AUPRC 0.77).
• Clinical Application: Produces predictions that correlate with clinical patient outcomes for predicting clinical responses.
• Drug Repurposing Potential: Analyzes pharmacogenomic datasets to identify and prioritize candidate drugs for repurposing across cancer types.

Scientific Applications:

• Drug response prediction: Predicts drug responses from integrated pharmacogenomic and multi-omics data to support precision oncology research.
• Drug repurposing prioritization: Ranks existing compounds for potential alternative cancer indications by analyzing integrated pharmacogenomic profiles.

Methodology:

Constructs kernel-based similarity matrices to integrate genomics, transcriptomics, epigenomics, chemical properties, and drug-target interactions, and trains deep neural networks on over 20,000 pan-cancer cell line-anticancer drug pairs; evaluations were performed on benchmark cancer cell line datasets and patient-derived cancer cell lines (AUC 0.84, AUPRC 0.77).