Machine Learning Prediction of Small Molecule Accumulation in Escherichia Coli Enhanced with Descriptor Statistics

Area of Science:

• Computational chemistry
• Microbiology
• Pharmacology

Background:

• Antibiotic resistance in Gram-negative bacteria is a major healthcare concern.
• Predicting small molecule accumulation is crucial for developing new antibacterial agents.
• Existing models may not fully capture the complex interactions within bacterial cells.

Purpose of the Study:

• To develop and optimize machine learning models for predicting small molecule accumulation in Gram-negative bacteria.
• To identify key molecular descriptors, including statistical ones, that enhance predictive accuracy.
• To gain insights into the physicochemical properties governing antibiotic uptake.

Main Methods:

• Utilized machine learning techniques combined with molecular dynamics simulations.
• Identified a minimal set of interpretable molecular descriptors.
• Incorporated statistical descriptors (e.g., dipole moment, minimum projection radius) into the models.
• Iteratively optimized model performance using metrics like accuracy and AUC.

Main Results:

• Statistical descriptors significantly improved model performance in predicting small molecule accumulation.
• Dipole moment and minimum projection radius descriptors were particularly impactful.
• The models demonstrated enhanced accuracy, precision, and AUC.
• The study provides insights into physicochemical properties essential for bacterial uptake.

Conclusions:

• Statistical moments beyond mean values are important for accurate predictive modeling.
• The developed approach offers a promising strategy for discovering and optimizing antibacterial agents.
• Findings are generalizable to Gram-negative bacteria, including *Escherichia coli*.