Identification of chemical features for improved outer membrane permeation in mycobacteria using machine learning.

The ability of compounds to permeate and accumulate in bacterial cells is a critical determinant of antibiotic efficacy. Better therapeutics are urgently needed for the human pathogen Mycobacterium tuberculosis, yet the cell envelope, including the mycobacterial outer membrane, represents a significant barrier for drug entry, and the chemical features governing permeation remain poorly understood. Here we used the bioorthogonal click chemistry-based PAC-MAN assay to profile mycomembrane permeation of 1,572 azide-tagged compounds in M. tuberculosis and the model organism M. smegmatis. Cheminformatics and machine learning identified chemical features associated with mycomembrane permeation, which in turn had predictive value in three molecule series. Chemical predictors of mycomembrane permeation include nitrogen-containing aromatic scaffolds, such as indole, which in some cases were associated with increased anti-M. tuberculosis activity. Our data suggest a rational framework for improving mycomembrane permeation and whole-cell activity of antibiotics targeting M. tuberculosis.