ContraDRG

ContraDRG predicts partial atomic charges of small molecules using machine learning models to reproduce charge assignments from computational chemistry sources for molecular modeling.

Key Features:

• Machine Learning Integration: Employs machine learning algorithms to predict partial atomic charges from molecular structures.
• Comparison with Existing Tools: Complements PRODRG and the Automated Topology Builder (ATB), both of which generate molecular topology files including partial atomic charges.
• Speed and Efficiency: Predicts partial charges in seconds compared with the hours or days required for ATB's quantum mechanical calculations.
• Model Training: Models are trained on partial charge data derived from PRODRG and ATB.
• Predictive Accuracy: Demonstrates predictive performance with R^2 up to 0.980 for ATB-derived charges and R^2 = 1.00 for PRODRG-derived charges.

Scientific Applications:

• Screening Projects: Enables rapid assignment of partial charges for large-scale molecular screening datasets.
• Research and Development: Supports biomedical and molecular research that requires accurate partial charge assignments for studying molecular interactions and properties.

Methodology:

Machine learning models were trained on partial charge datasets from PRODRG and ATB, with reported predictive performance of R^2 = 0.980 for ATB-derived charges and R^2 = 1.00 for PRODRG-derived charges.