HPDAF: A practical tool for predicting drug-target binding affinity using multimodal features.

Area of Science:

• Computational chemistry
• Drug discovery and design
• Bioinformatics

Background:

• Accurate prediction of drug-target binding affinity is essential for efficient drug discovery.
• Current computational methods struggle to integrate diverse molecular features effectively.
• There is a need for advanced tools to improve binding affinity prediction accuracy.

Purpose of the Study:

• To introduce HPDAF, a multimodal deep learning tool for enhanced drug-target binding affinity prediction.
• To develop a method that effectively integrates protein sequences, drug molecular graphs, and binding pocket structural data.
• To improve the accuracy and practical applicability of computational drug discovery tools.

Main Methods:

• HPDAF utilizes a multimodal deep learning approach.
• It integrates protein sequences, drug molecular graphs, and protein-binding pocket structural data.
• A hierarchical attention mechanism combines these features for dynamic emphasis on relevant information.

Main Results:

• HPDAF demonstrates superior predictive performance on benchmark datasets (CASF-2016, CASF-2013).
• The model effectively integrates diverse biochemical information for improved accuracy.
• Consistent superior performance compared to state-of-the-art methods was observed.

Conclusions:

• HPDAF offers enhanced accuracy for drug-target binding affinity predictions.
• The tool's practical applicability benefits medicinal chemists in drug design and virtual screening.
• HPDAF represents a valuable advancement in computational drug discovery.