Machine Learning and Deep Learning Models for Predicting Noncovalent Inhibitors of AmpC β-Lactamase.

Area of Science:

• Medicinal Chemistry
• Computational Biology
• Drug Discovery

Background:

• Antibiotic resistance (AR) is a growing global health threat, driven by bacterial adaptation mechanisms like beta-lactamase synthesis.
• Class C beta-lactamase (AmpC) confers resistance to crucial beta-lactam antibiotics, necessitating the development of novel therapeutic strategies.

Purpose of the Study:

• To develop accurate machine learning (ML) and deep learning (DL) models for predicting noncovalent inhibitors of AmpC beta-lactamase.
• To identify novel chemical entities with potential as AmpC inhibitors to address urgent treatment needs.

Main Methods:

• Utilized large compound datasets for training and validation.
• Developed and compared Support Vector Machine (SVM), Random Forest (RF), and Feed-Forward Neural Network (FFNN) classification models.
• Analyzed physicochemical properties and predicted binding modes of identified inhibitors.

Main Results:

• Achieved cross-validation accuracies ranging from 80% to 82% for individual models.
• Combined ML/DL models reached an overall accuracy of 83% in predicting noncovalent AmpC inhibitors.
• Identified key physicochemical characteristics and binding modes relevant for inhibitor design.

Conclusions:

• The developed ML/DL models provide a robust and accurate platform for the virtual screening and identification of novel noncovalent AmpC inhibitors.
• These predictive models are valuable tools for accelerating the discovery of new solutions against the escalating challenge of beta-lactamase-mediated antibiotic resistance.