ML-PLA: Enhancing Protein-Ligand Binding Affinity Prediction with Microenvironment and Long-Range Interaction-Aware...

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Accurate prediction of protein-ligand binding affinity (PLA) is crucial for drug discovery.
• Existing graph-based models often use shallow feature fusion and fixed distance thresholds, limiting their ability to capture complex biological interactions.

Purpose of the Study:

• To develop a novel method, ML-PLA, for enhanced prediction of protein-ligand binding affinity.
• To address limitations in current models regarding microenvironment representation and long-range interaction modeling.

Main Methods:

• Utilized a heterogeneous graph neural network to aggregate sequence and structural information for protein microenvironment modeling.
• Employed a vector quantized-variational autoencoder for diverse and chemically meaningful microenvironment representations.
• Implemented a multihead attention mechanism to project atoms into virtual atoms, capturing long-range interactions without oversmoothing.

Main Results:

• ML-PLA demonstrated significant effectiveness in predicting protein-ligand binding affinity.
• The method showed robust generalization capabilities on benchmark datasets (CASF-2016 and CASF-2013).
• Outperformed state-of-the-art methods in binding affinity prediction.

Conclusions:

• ML-PLA offers a superior approach to modeling protein microenvironments and long-range interactions for accurate PLA prediction.
• The method holds promise for advancing lead compound identification in drug discovery.