Combined substituent number utilized machine learning for the development of antimicrobial agent

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Machine learning (ML) offers potential for advancing antimicrobial agent development.
• Effective ML application requires molecular descriptors balancing informativeness and computational feasibility.
• Existing descriptors can be computationally intensive or lack predictive validity.

Purpose of the Study:

• To develop and evaluate a novel molecular descriptor for machine learning in antimicrobial drug discovery.
• To assess the predictive performance of this descriptor for 4-quinolone antimicrobial agents.
• To demonstrate the utility of the descriptor in generating and screening large virtual compound libraries.

Main Methods:

• A combined substituent number (CSN) descriptor was developed, focusing on substituent type and position in 4-quinolone structures.
• Machine learning models were trained and validated using the CSN descriptor for 11,879 known compounds.
• A large virtual library of over 32 million compounds was generated by substituent recombination.

Main Results:

• The CSN-based ML model achieved a coefficient of determination of 0.719 for training and 0.519 for validation.
• The CSN descriptor enabled the efficient construction of a large, structurally consistent virtual library.
• Experimental validation confirmed the predictive accuracy of the model for E. coli growth inhibition.

Conclusions:

• The CSN descriptor provides a computationally efficient and valid approach for ML-driven antimicrobial drug discovery.
• This method facilitates rapid virtual screening and identification of potential antimicrobial candidates.
• The study highlights the practical application of simplified molecular descriptors in accelerating drug development pipelines.