MechGate: combining mechanism-derived measures of permeation and efflux to screen small molecule drugs targeting Escherichia coli

Abstract

Most antibiotic molecules must cross the cell membrane and escape efflux pumps before hitting their targets. For de novo antibiotic design, very few of the in-silico-generated drug molecules can successfully pass these two steps. Existing evaluations of generated molecules do not separate these steps. Therefore, we introduce MechGate, a mechanism-resolved screening tool evaluating small drug molecules against E. coli, measuring permeation, efflux and antibacterial activity. After SMILES input, five measures are reported side by side: two for permeation via porins, one for permeation via the phospholipid membrane, one for efflux and one for activity. Each measure is read on its own, so a molecule that fails can be traced to the step that stops it. All features are taken from papers that state their mechanism rather than selected by importance ranking. On an external set of marketed antibacterials the measures give a precision of 0.789 to 0.873 against baselines near 0.44, and combining them the way the mechanism predicts raises precision to 0.92. We also find that the four public source libraries are separable from one another using physicochemical features alone; every measure here therefore draws both classes from a single library. All code is open-source and requires no commercial software.