Alphappimi: a comprehensive deep learning framework for predicting PPI-modulator interactions.

Area of Science:

• Computational Biology
• Drug Discovery
• Bioinformatics

Background:

• Protein-protein interactions (PPIs) are crucial for biological processes, and their dysregulation is linked to diseases.
• Identifying modulators that target PPIs and their interfaces is a key therapeutic strategy.
• Traditional methods struggle to identify PPI modulators, especially for targets lacking known active compounds.

Purpose of the Study:

• To develop a deep learning framework, AlphaPPIMI, for predicting protein-protein interaction modulator (PPIMI) interactions.
• To specifically target PPI interfaces for modulator discovery.
• To create robust benchmark datasets for evaluating PPIMI prediction methods.

Main Methods:

• Integrated multimodal molecular features (Uni-Mol2), protein representations (ESM2, ProTrans), and PPI structural characteristics (PFeature).
• Employed a specialized cross-attention architecture for fusing diverse molecular representations.
• Utilized Conditional Domain Adversarial Networks (CDAN) for enhanced cross-domain generalization.

Main Results:

• AlphaPPIMI demonstrated superior performance in PPIMI prediction compared to existing methods.
• The framework effectively learned associations between PPI targets and modulators.
• Achieved robust cross-domain generalization across diverse protein families.

Conclusions:

• AlphaPPIMI offers a powerful computational tool for prioritizing candidate PPI modulators.
• The framework shows promise for the discovery of targeted PPI therapeutics, particularly those acting on protein-protein interfaces.
• This work advances the field of computational drug discovery for complex protein targets.