P2MAT: A machine learning (ML) driven software for Property Prediction of MATerial.

Area of Science:

• Chemistry
• Materials Science
• Drug Discovery

Background:

• Accurate melting point prediction is crucial but challenging for pure molecules.
• Traditional methods like group contribution (GC) have limitations due to complex structure-property relationships.

Purpose of the Study:

• To develop a data-driven machine learning (ML) approach for predicting melting points of organic compounds.
• To leverage both 2D and 3D molecular descriptors for enhanced prediction accuracy.
• To identify key molecular features influencing melting point predictions through SHAP explainability.

Main Methods:

• Utilized a machine learning (ML) model trained on molecular descriptors.
• Incorporated both 2D and 3D molecular descriptors into the ML model.
• Employed SHAP (SHapley Additive exPlanations) for feature importance analysis.

Main Results:

• Machine learning models significantly improved the accuracy of melting point predictions compared to traditional methods.
• Identified top molecular features that influence melting point predictions.
• Developed the P2MAT application for predicting melting and boiling points from SMILES strings.

Conclusions:

• The developed ML approach offers a robust and accurate method for predicting melting points.
• The P2MAT application provides a user-friendly tool for the scientific community.
• This work advances predictive chemistry with practical applications in materials science and drug discovery.