DynamicDTA: Drug-Target Binding Affinity Prediction Using Dynamic Descriptors and Graph Representation.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Predicting drug-target binding affinity (DTA) is crucial for identifying therapeutic candidates.
• Existing DTA models often neglect protein dynamics, limiting accuracy.
• Protein conformational flexibility plays a key role in binding interactions.

Purpose of the Study:

• To introduce DynamicDTA, a deep learning framework for enhanced DTA prediction.
• To incorporate both static and dynamic protein features into DTA modeling.
• To improve the accuracy and reliability of DTA predictions.

Main Methods:

• DynamicDTA utilizes drug sequences, protein sequences, and dynamic descriptors as input.
• Graph convolutional networks process drug molecular graphs; dilated convolutions encode protein sequences.
• Dynamic descriptors are processed via multi-layer perceptrons, fused with static protein features using cross-attention, and integrated via a tensor fusion network.

Main Results:

• DynamicDTA achieved at least a 3.4% improvement in RMSE score compared to seven state-of-the-art methods across three datasets.
• The model demonstrated reliability and biological relevance in predicting novel drugs for HIV-1.
• Visualization of docking complexes further validated the model's predictions.

Conclusions:

• DynamicDTA offers a significant advancement in DTA prediction by incorporating protein dynamics.
• The framework enhances accuracy and provides biologically relevant insights for drug discovery.
• The source code is publicly available, facilitating further research and application.