MDDTA: A Drug Target Binding Affinity Prediction Method Based on Molecular Dynamics Simulation Data Enhancement.

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Deep learning models for drug target binding affinity (DTA) prediction are crucial for efficient drug screening.
• Current methods often overlook the conformational dynamics of drug-target complexes, limiting their ability to capture subtle affinity variations.

Purpose of the Study:

• To develop a deep learning model that accounts for conformational dynamics in DTA prediction.
• To enhance the accuracy and applicability of computational drug screening.

Main Methods:

• Construction of MD-PDBbind, a dataset utilizing molecular dynamics (MD) simulations.
• Development of the MDDTA model featuring the FAFormer architecture for geometric learning and a dynamic-aware loss function.
• Evaluation using the CASF-2016 dataset and a drug screening campaign for SARS-CoV-2 compounds.

Main Results:

• MDDTA achieved excellent scoring and ranking performance on the CASF-2016 benchmark.
• A case study validated the benefit of incorporating dynamic information into DTA prediction.
• Screening of 70 SARS-CoV-2 compounds identified five promising candidates, with literature validation.

Conclusions:

• The MDDTA model effectively integrates conformational dynamics for improved DTA prediction.
• This approach enhances the potential of computational methods in practical drug discovery and development.
• The identified SARS-CoV-2 compounds warrant further investigation for therapeutic applications.