KEPLA: A Knowledge-Enhanced Deep Learning Framework for Accurate Protein-Ligand Binding Affinity Prediction

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for efficient drug discovery.
• Current deep learning methods often neglect valuable biochemical knowledge, focusing primarily on structural features.
• This limitation hinders the full potential of computational approaches in identifying drug candidates.

Purpose of the Study:

• To introduce KEPLA, a novel deep learning framework designed to improve protein-ligand binding affinity prediction.
• To explicitly integrate prior biochemical knowledge, including Gene Ontology and ligand properties, into the prediction model.
• To enhance the accuracy and interpretability of binding affinity predictions.

Main Methods:

• KEPLA utilizes protein sequences and ligand molecular graphs as input.
• It employs two key objectives: aligning global representations with knowledge graph relations and using cross-attention for joint embeddings.
• The framework leverages Gene Ontology and ligand properties to capture biochemical insights.

Main Results:

• KEPLA consistently outperforms state-of-the-art baseline methods on two benchmark datasets.
• The model demonstrates strong performance in both in-domain and cross-domain prediction scenarios.
• Interpretability analyses reveal insights into the model's predictive mechanisms via knowledge graph relations and attention maps.

Conclusions:

• KEPLA represents a significant advancement in predicting protein-ligand binding affinity by incorporating biochemical knowledge.
• The framework offers a more comprehensive approach compared to structure-only methods.
• KEPLA's ability to provide interpretable results enhances its utility in drug discovery research.