Interpretable machine learning models for QSAR-based prediction of anti-Salmonella typhi activity

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Machine learning

Background:

• Multidrug-resistant Salmonella typhi poses a significant global health threat.
• Novel therapeutic strategies are urgently needed to combat resistant bacterial infections.

Purpose of the Study:

• To develop a robust machine learning (ML)-based quantitative structure-activity relationship (QSAR) model.
• To identify potential drug candidates effective against multidrug-resistant Salmonella typhi.

Main Methods:

• A curated ChEMBL-derived dataset was used, achieving a high modelability score (MODI = 0.89).
• A hybrid feature selection workflow identified 20 chemically interpretable molecular descriptors.
• Eight diverse ML classifiers were trained and benchmarked, including Support Vector Machine (SVM).

Main Results:

• The SVM model demonstrated the highest performance on the hold-out test set.
• Achieved a Matthews Correlation Coefficient (MCC) of 0.61 and ROC-AUC of 0.90.
• Successfully identified potential drug candidates for anti-S. typhi activity.

Conclusions:

• Rigorous ML-QSAR modeling provides a reliable framework for drug discovery.
• This approach enables efficient virtual screening and prioritization of novel anti-S. typhi agents.
• Facilitates the development of new treatments against resistant bacterial pathogens.