SAnDReS

SAnDReS performs statistical analysis and machine-learning-based evaluation of protein–ligand docking results to improve selection of docking protocols and prediction of ligand-binding affinity.

Key Features:

• Integration with Docking Programs: Integrates with AutoDock4 and Molegro Virtual Docker (MVD) to execute docking simulations.
• Statistical Analysis of Docking Results: Performs detailed statistical evaluation of multiple docking protocols to enable selection of optimal strategies for specific protein systems.
• Machine Learning Integration: Implements supervised machine-learning methods, including convolutional neural networks and random forests, trained on atomic coordinates of protein–ligand complexes to predict ligand-binding affinity and develop targeted scoring functions for proteins such as cyclin-dependent kinase and HIV-1 protease.
• Workflow Integration and Protocol Variants: Combines docking steps into a unified workflow and accounts for presence or absence of water molecules, producing up to 32 different docking protocol variants when combined with MVD.
• Ensemble Support: Supports simulations of ensembles of crystallographic structures for which ligand-binding affinity data are available.

Scientific Applications:

• Binder Identification: Identification of potential new binders to protein targets through docking and affinity prediction.
• Protocol Selection and Validation: Selection and validation of docking protocols tailored to specific protein systems.
• Ensemble-Based Affinity Analysis: Simulation and analysis of ensembles of crystallographic structures linked to ligand-binding affinity data.
• Targeted Scoring Function Development: Development of protein-specific scoring functions, exemplified for cyclin-dependent kinase and HIV-1 protease.

Methodology:

SAnDReS runs docking simulations via AutoDock4 and Molegro Virtual Docker (MVD) using various search algorithms and scoring functions; it evaluates multiple docking protocols including presence or absence of water (32 protocol variants with MVD) and applies supervised machine learning—convolutional neural networks and random forests—trained on atomic coordinates of protein–ligand complexes to predict binding affinity and derive targeted scoring functions.