Machine Learning Pipeline for Molecular Property Prediction Using ChemXploreML.

Area of Science:

• Cheminformatics
• Computational Chemistry
• Machine Learning

Background:

• Accurate prediction of molecular properties is crucial for drug discovery and materials science.
• Existing tools often lack flexibility in integrating diverse machine learning models and molecular representations.
• Need for user-friendly platforms that democratize access to advanced cheminformatics techniques.

Purpose of the Study:

• To introduce ChemXploreML, a modular desktop application for machine learning-based molecular property prediction.
• To enable researchers to customize prediction pipelines by integrating various molecular embedding techniques and machine learning algorithms.
• To demonstrate the framework's utility and performance using established molecular property datasets.

Main Methods:

• Implemented ChemXploreML with a flexible architecture supporting diverse embedding methods and ML algorithms.
• Evaluated Mol2Vec and VICGAE (Variance-Invariance-Covariance regularized GRU Auto-Encoder) embeddings combined with Gradient Boosting, XGBoost, CatBoost, and LightGBM.
• Validated the framework on predicting melting point, boiling point, vapor pressure, critical temperature (CT), and critical pressure using CRC Handbook data.

Main Results:

• Achieved excellent predictive performance for well-distributed properties, with R-squared values up to 0.93 for critical temperature.
• Mol2Vec embeddings (300D) showed slightly higher accuracy, while VICGAE embeddings (32D) offered comparable performance with improved computational efficiency.
• Demonstrated the framework's capability in automating chemical data preprocessing, model optimization, and performance analysis.

Conclusions:

• ChemXploreML provides a flexible and accessible platform for customized molecular property prediction.
• The modular design facilitates easy integration of novel embedding techniques and ML algorithms.
• The application empowers researchers, from beginners to advanced users, to leverage sophisticated machine learning in cheminformatics.