ContraDRG: Automatic Partial Charge Prediction by Machine Learning.

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Accurate partial atomic charges are crucial for molecular modeling and simulations in drug discovery.
• Existing methods like PRODRG and Automated Topology Builder (ATB) provide these charges but involve computationally intensive calculations.
• The need for faster, efficient methods for charge prediction in large-scale molecular screening is significant.

Purpose of the Study:

• To develop a machine learning-based software, ContraDRG, for rapid prediction of partial atomic charges.
• To emulate computationally complex predictions from PRODRG and ATB using machine learning.
• To provide a web server for automated partial charge assignment for user-specified molecules.

Main Methods:

• Utilized machine learning algorithms to train models based on existing PRODRG and ATB predictions.
• Developed ContraDRG software to predict partial charges by emulating computationally intensive calculations.
• Compared the predictivity of ContraDRG against PRODRG and ATB using statistical methods.

Main Results:

• ContraDRG accurately predicts partial charges, achieving an R value up to 0.980 for ATB-derived charges and 1.00 for PRODRG-derived charges.
• ContraDRG performs predictions in seconds, drastically reducing the time compared to hours or days required by ATB.
• The software is provided as a web server for easy access and automated charge assignment.

Conclusions:

• ContraDRG offers a computationally efficient and accurate alternative for predicting partial atomic charges.
• The speed of ContraDRG makes it suitable for screening large numbers of molecules in drug discovery projects.
• Machine learning effectively emulates complex quantum mechanical calculations for partial charge prediction.