Machine learning-based quantitative structure-activity relationship model for antibiotic prediction and discovery

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Antimicrobial resistance (AMR) poses a significant global health threat.
• Traditional antibiotic screening is slow and costly, hindering the development of new treatments.
• Machine learning and QSAR offer a faster, more efficient approach to identifying potential drug candidates.

Purpose of the Study:

• To develop and validate a predictive model for antibacterial activity using machine learning and QSAR.
• To accelerate the discovery of novel antibiotic compounds.
• To identify key chemical descriptors influencing antibacterial efficacy.

Main Methods:

• Collected compound data from PubChem, curating active and inactive datasets.
• Employed machine learning algorithms: Decision Tree (DT), Support Vector Machine (SVM), and Naïve Bayesian (NB).
• Predicted antibacterial activity based on Simplified Molecular Input Line Entry System (SMILES) and evaluated model performance.

Main Results:

• DT and SVM models demonstrated high predictive performance (accuracy, precision, sensitivity, AUC-ROC > 0.90), outperforming NB.
• Electrotopy and β-lactam structure descriptors were identified as key predictors.
• The model showed high accuracy across different antibiotic classes.

Conclusions:

• A robust QSAR-integrated machine learning model can effectively predict antibacterial activity.
• This approach significantly accelerates the identification of potential novel antibiotics.
• The predictive model is accessible via a web application for broader use.