A multi-modal drug-gene interaction prediction framework for recommending drugs targeting epigenetic cancer...

Area of Science:

• Computational biology
• Genomics
• Drug discovery

Background:

• Drug-target interaction analysis is crucial for cancer therapy and precision medicine.
• Current methods overlook disease-specific molecular and functional contexts.
• There is a need for advanced computational frameworks to predict drug-gene interactions effectively.

Purpose of the Study:

• To develop a multi-modal, gene-centric, graph-based chemogenomic framework (M-DGI) for predicting drug-gene interactions.
• To integrate chemical structure, DNA sequence, and pathway information for enhanced prediction accuracy.
• To identify novel drug candidates targeting cancer biomarkers for personalized treatment.

Main Methods:

• Developed M-DGI, a graph-based chemogenomic framework.
• Utilized multi-modal gene embeddings (DNA sequence, pathway information) and chemical structure-based drug embeddings.
• Employed GraphSAGE, an inductive graph neural network, on a bipartite drug-gene interaction graph.

Main Results:

• M-DGI achieved high performance in drug-gene interaction prediction.
• Demonstrated superior performance over existing techniques with an AUC of 91.85% and average precision of 92.84%.
• Successfully recommended potential drug candidates targeting pan-cancer and cancer-specific biomarkers.

Conclusions:

• M-DGI offers a powerful approach for drug-gene interaction prediction, advancing precision cancer therapy.
• The framework's ability to integrate multi-modal data enhances the identification of targeted therapies.
• M-DGI supports personalized treatment strategies by identifying effective drug candidates for specific cancer biomarkers.