DLA-Ranker

DLA-Ranker evaluates protein docking conformations to identify near-native protein complex structures by analyzing local environments around interfacial residues.

Key Features:

• Deep learning framework: DLA-Ranker employs 3D convolutional neural networks to process volumetric representations of protein interfaces.
• Locally oriented cubes: The protein interface is represented using locally oriented cubes that capture local geometric information, neighboring atoms, and variations in solvent accessibility around interfacial residues.
• Conformation discrimination: The method discriminates near-native conformations from incorrect ones within large ensembles generated by molecular docking.
• Benchmark performance: DLA-Ranker was evaluated on three docking benchmarks, each comprising half a million acceptable and incorrect conformations, and shows superior or competitive performance versus other deep learning-based scoring functions.
• Alternative interface discovery: The approach can identify alternative interfaces within protein complexes.

Scientific Applications:

• Protein–protein interaction analysis: Identification of near-native conformations to improve the characterization of protein–protein interfaces.
• Docking scoring and model selection: Ranking and selection of biologically relevant models from large docking ensembles in molecular modeling workflows.
• Drug discovery: Exploration of interaction sites and alternative interfaces that may be relevant for drug design.
• Structural biology: Investigation of alternative interfaces and interaction mechanisms relevant to understanding disease processes.

Methodology:

DLA-Ranker applies 3D convolutional neural networks to locally oriented cubes representing protein interfaces to analyze local geometric features and solvent accessibility, and the model was trained and evaluated on three docking benchmarks each comprising half a million acceptable and incorrect conformations.