Convex-PL-R

Convex-PL-R predicts binding poses and affinities for structure-based virtual screening by incorporating conformational flexibility and entropic modeling to reduce bias toward larger binding interfaces.

Key Features:

• Conformational Flexibility: Incorporates conformational sampling to allow flexible ligand conformations during binding-pose and affinity prediction.
• Novel Dataset Training: Trained on a dataset enriched with cofactors to capture diverse molecular interactions.
• Physical Model with Entropic Terms: Adds entropic terms to the physical scoring model to penalize the bias toward larger binding interfaces.
• Machine Learning Integration: Parameterized by solving a classification problem followed by regression using affinity and structural data.
• CASF Benchmark Analysis: Evaluates scoring-function performance using CASF benchmarks to identify and address interface-size bias.
• Integration with VinaCPL: Integrates with VinaCPL (a version of AutoDock Vina) to serve as an advanced scoring function in docking workflows.

Scientific Applications:

• Structure-based virtual screening: Predicts binding poses and affinities to prioritize candidate compounds in drug discovery.
• Docking scoring for AutoDock Vina/VinaCPL: Provides an alternative scoring function for docking runs performed with VinaCPL.
• Modeling systems with cofactors: Supports prediction in complexes that include cofactors due to targeted dataset enrichment.
• Bias mitigation in scoring functions: Reduces tendency to favor larger binding interfaces through entropic penalties.

Methodology:

Analyzes scoring-function performance in CASF benchmarks, incorporates entropic penalties into a physical scoring model, implements conformational sampling for ligand flexibility, and parameterizes the model via machine learning by solving a classification task followed by regression using affinity and structural data.