BIPSPI+

BIPSPI+ predicts partner-specific binding sites in protein-protein interactions using XGBoost-based machine learning to identify interface residues and support 3D quaternary structure modeling.

Key Features:

• Advanced Machine Learning: Uses the XGBoost algorithm to predict specific residues that form protein-protein interfaces.
• Curated Training Datasets: Trained on carefully curated datasets selected to represent homo- and hetero-complex interactions.
• Enhanced Prediction Capabilities: Provides improved partner-specific interface prediction accuracy relative to the previous BIPSPI version.
• Specialized Modes: Includes a sequence-structure prediction mode and specific handling for hetero- or homo-complexes to tailor predictions to different complex types.
• Guided Docking Tool: Computes 3D quaternary structure poses using predicted interfaces to aid modeling of protein assemblies.

Scientific Applications:

• Interface residue identification: Delineates residues involved in protein-protein interactions for molecular-level analysis.
• Drug design: Supports identification of binding-site residues relevant for small-molecule or biologic targeting.
• Functional annotation: Assists in assigning functional roles to proteins based on predicted interaction interfaces.
• Structural biology: Aids in modeling and interpretation of 3D quaternary structures of protein complexes.
• Pathway elucidation: Facilitates mapping of interaction interfaces within biological pathways.

Methodology:

Training XGBoost machine learning models on curated datasets representing various protein interaction types, with integration of a sequence-structure prediction mode and use of predicted interfaces to guide computation of 3D docking poses.