M-polynomial driven machine learning models for predicting physicochemical properties of antibiotics.

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Drug Discovery

Background:

• Accurate prediction of drug compound physicochemical properties is essential for developing safe and effective antibiotics.
• Existing methods may lack the precision and robustness required for real-world drug development.

Purpose of the Study:

• To develop and evaluate advanced machine learning models for predicting physicochemical properties of drug compounds.
• To assess the generalization capability and robustness of Support Vector Regression (SVR) and Random Forest (RF) models.

Main Methods:

• Utilized M-Polynomials and physicochemical descriptors as input features for machine learning models.
• Implemented and compared basic SVR (SVR-Basic), optimized SVR (SVR-Tuned), and Random Forest (RF) models.
• Assessed model performance using R2, MSE, RMSE, MAE, and detailed residual analyses (MR, Std Residual, IQR).

Main Results:

• The machine learning models demonstrated improved predictive accuracy compared to previous studies.
• Feature importance analysis and ablation studies provided insights into descriptor contributions and model stability.
• Visual comparisons confirmed robust model behavior on training and test datasets.

Conclusions:

• The proposed machine learning framework offers a robust and reliable approach for predicting drug compound properties.
• This methodology enhances the efficiency and accuracy of antibiotic drug development.
• The study highlights the practical applicability of Python-based machine learning in pharmaceutical research.