AlphaDTA: integrating AlphaFold3 embeddings and 3D complex structures for drug-target binding affinity prediction

Area of Science:

• Computational Biology
• Drug Discovery
• Structural Bioinformatics

Background:

• Accurate prediction of drug-target binding affinity is crucial for structure-based drug discovery.
• Existing methods are limited by the scarcity of experimentally determined protein-ligand complex structures.
• Advances in biomolecular structure prediction, like AlphaFold3, offer a potential solution.

Purpose of the Study:

• To introduce AlphaDTA, a novel framework for drug-target binding affinity prediction.
• To leverage AlphaFold3-predicted structures and embeddings to overcome data limitations.
• To evaluate AlphaDTA's performance on structurally nonredundant benchmarks and its utility in drug repurposing.

Main Methods:

• AlphaDTA processes AlphaFold3 single and pair embeddings for interaction patterns and relational features.
• A 3D geometric encoder generates structural embeddings from AlphaFold3-predicted structures.
• Adaptive fusion integrates single, pair, and structural embeddings for affinity prediction.

Main Results:

• AlphaDTA achieved state-of-the-art or competitive performance on benchmarks with reduced structural overlap.
• The framework demonstrates improved generalization on structurally diverse datasets.
• In a case study, AlphaDTA identified a known drug and suggested a repurposing candidate for cystic fibrosis.

Conclusions:

• AlphaDTA enables accurate drug-target binding affinity prediction without experimental complex structures.
• Integration of diverse AlphaFold3-derived embeddings via adaptive fusion enhances predictive power.
• The framework shows promise for structure-based drug discovery and targeted drug repurposing.