Machine Learning Models and Structure-Based Antibacterial Drug Discovery of the Key ABC Transporter Maltose-Binding...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Multidrug-resistant gram-negative bacterial infections pose a significant global health threat.
• Developing novel antibacterial agents is crucial to combatting resistance.
• Machine learning-assisted quantitative structure-activity relationships (ML-QSAR) offer a promising strategy for efficient drug discovery.

Purpose of the Study:

• To develop robust ML-QSAR models for predicting the inhibitory activity of quinoline-based compounds against MsbA.
• To identify novel, potent MsbA inhibitors for combating multidrug-resistant gram-negative bacteria.
• To create a user-friendly web application for predicting MsbA inhibitory activity.

Main Methods:

• Utilized Genetic Function Approximation (GFA), Support Vector Machine (SVM), and Artificial Neural Network (ANN) for ML-QSAR model development.
• Employed eight molecular descriptors and 279 molecular fingerprints to build predictive models.
• Validated model robustness through internal, external, and applicability domain analyses.

Main Results:

• The molecular fingerprint-based SVM model demonstrated superior performance with R² = 0.9981 and q² = 0.7584.
• Identified compounds 31 and 40 as highly active, leading to the generation of 62 new compounds.
• Generated three novel quinoline-based inhibitors (M28, N7, N23) with excellent predicted bioactivity, binding affinity, and pharmacokinetic profiles.

Conclusions:

• Developed and validated effective ML-QSAR models for predicting MsbA inhibitory activity.
• Successfully identified promising novel quinoline-based antibacterial drug candidates.
• Launched the MsbA-Pred web application to facilitate further research and drug discovery efforts.