Random Forest Modeling to Predict Small Molecule Accumulation in Gram-Negative Bacteria

Area of Science:

• Microbiology
• Computational Chemistry
• Drug Discovery

Background:

• The development pipeline for Gram-negative-active antibiotics is limited due to complex bacterial membranes and poor understanding of compound accumulation.
• Physicochemical properties influencing antibiotic permeation, efflux evasion, and accumulation are critical for Gram-negative pathogens.

Purpose of the Study:

• To describe a workflow using random forest modeling to analyze physicochemical trends linked to small molecule accumulation in Gram-negative bacteria.
• To provide guidelines for small molecule accumulation based on machine learning analysis.

Main Methods:

• Utilized unbiased accumulation assays and advanced chemical descriptors.
• Applied machine learning algorithms, specifically random forest modeling, to correlate physicochemical properties with compound accumulation.
• Analyzed accumulation data in *E. coli* and *P. aeruginosa*.

Main Results:

• Identified key physicochemical trends associated with small molecule permeation and efflux liabilities in Gram-negative pathogens.
• Developed guidelines for enhancing small molecule accumulation in *E. coli* and *P. aeruginosa*.
• Random forest modeling proved effective in processing large datasets and providing interpretable insights.

Conclusions:

• Random forest modeling offers significant advantages for identifying physicochemical trends in antibiotic accumulation.
• This approach generates valuable hypotheses for the study of small molecule accumulation in Gram-negative bacteria.
• The developed guidelines have led to the discovery of new Gram-negative-active antibiotics.