A data-driven machine learning approach for discovering potent LasR inhibitors.

Area of Science:

• Computational chemistry and drug discovery
• Machine learning applications in pharmacology
• Antimicrobial resistance research

Background:

• Multidrug-resistant Pseudomonas aeruginosa poses a significant global health threat.
• Limited development of new antibiotics necessitates alternative therapeutic strategies.
• Targeting the las quorum sensing (QS) system is a promising approach against P. aeruginosa.

Purpose of the Study:

• To develop a machine learning-based drug prediction algorithm for identifying potent LasR inhibitors.
• To screen chemical databases for novel compounds targeting the P. aeruginosa las QS system.
• To validate potential inhibitors through computational analyses.

Main Methods:

• Development and optimization of a Multilayer Perceptron (MLP) algorithm with AdaBoostM1.
• Evaluation of model performance using 5-fold cross-validation and test sets.
• Virtual screening of the Enamine database and subsequent molecular docking, Molecular Dynamics, MM-GBSA, and Free Energy Landscape analyses.

Main Results:

• The best MLP model achieved 90.7% accuracy, an AUC of 0.95, and an MCC of 0.81 in discriminating LasR inhibitors.
• Virtual screening identified several top-ranked compounds with superior predicted ligand-binding affinities compared to naringenin.
• Five out of six top hits were predicted as potent LasR inhibitors with potential therapeutic applications.

Conclusions:

• The study presents the first assessment of an MLP-based Quantitative Structure-Activity Relationship (QSAR) model for discovering LasR inhibitors.
• The developed model effectively identifies potential drug candidates against P. aeruginosa.
• The identified compounds represent promising leads for developing new treatments for P. aeruginosa infections.