PI2PE

PI2PE predicts solvent accessibility and interface residues in protein–protein and protein–DNA complexes to support structural interpretation of binding sites.

Key Features:

• WESA (Web-based Ensemble Solvent Accessibility): Predicts residue solvent accessibility from sequence using ensemble techniques that consider amino acid side-chain physical properties.
• cons-PPISP (Consensus Protein-Protein Interface Site Predictor): Identifies protein–protein interface residues using sequence profiles and solvent accessibility of spatially neighboring surface residues as inputs to neural network models and combines multiple neural-network predictions in a consensus approach.
• cons-PPISP training and performance: Neural networks were trained on 1156 nonhomologous protein chains and reported ~80% accuracy with 51% coverage of native interface residues.
• DISPLAR (Displacement-based Prediction): Predicts interface residues by analyzing structural changes between bound and unbound states using neighbor lists and solvent exposure measures that are relatively insensitive to conformational alterations, with demonstrated high accuracy across test sets.
• Predictor integration: Integrates WESA, cons-PPISP, and DISPLAR outputs to provide combined predictions of solvent accessibility and interface residues.
• PDB-derived training data: Uses neural-network–based approaches trained on datasets derived from the Protein Data Bank.

Scientific Applications:

• Structural modeling of complexes: Supports construction of structural models for multicomponent protein–DNA complexes and protein–protein interactions.
• DNA-binding site prediction: Predicts DNA-contacting residues from protein structures alone with reported 76% accuracy.
• Experimental complementarity: Provides interface residue predictions that have been consistent with NMR chemical shift perturbation data for characterizing protein–protein interfaces.

Methodology:

Uses ensemble solvent-accessibility prediction (WESA), consensus neural-network models informed by sequence profiles and neighboring-surface residue solvent accessibility (cons-PPISP), and displacement-based analysis of neighbor lists and solvent exposure between bound and unbound states (DISPLAR); neural networks were trained on Protein Data Bank–derived datasets.