MMPD-DTA: Integrating Multi-Modal Deep Learning with Pocket-Drug Graphs for Drug-Target Binding Affinity Prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate prediction of drug-target binding affinity (DTA) is essential for efficient drug discovery.
• Existing methods struggle to fully capture both global and local information, and effectively model pocket features.
• Understanding target-drug interactions at atomic and global levels is key to improving DTA prediction.

Purpose of the Study:

• To propose a novel multimodal deep learning model, MMPD-DTA, for enhanced DTA prediction.
• To address limitations in considering global/local information and modeling pocket features in DTA prediction.
• To develop a model that integrates diverse data modalities for comprehensive target-drug interaction analysis.

Main Methods:

• Developed the MMPD-DTA model, integrating graph and sequence data from targets, pockets, and drugs.
• Introduced a novel pocket-drug (PD) graph to model atomic interactions within and between targets and drugs.
• Employed GraphSAGE for PD graph learning, transformers for target sequences, and Graph Isomorphism Network for drug graphs, followed by MLP prediction.

Main Results:

• MMPD-DTA demonstrated superior performance compared to baseline methods on three real-world datasets.
• Ablation studies validated the contribution and effectiveness of individual components within the MMPD-DTA model.
• The model successfully integrated multimodal representations for accurate binding affinity prediction.

Conclusions:

• The MMPD-DTA model offers a significant advancement in predicting drug-target binding affinity.
• Integrating multimodal data and a novel PD graph effectively captures complex target-drug interactions.
• The proposed approach provides a robust framework for computational drug discovery research.