Predicting permeability and efflux of antibiotic compounds
Year: 2026
Type: article
Abstract: Whether a compound penetrates Gram-negative bacteria and escapes efflux remains a major bottleneck in antibiotic discovery, and existing predictive rules and models are built largely from E. coli data, limited strain panels, or proxy readouts, leaving it unclear which accumulation determinants transfer across pathogens. This pilot proposes a standardised LC-MS/MS benchmark of intracellular compound accumulation measured in E. coli, Acinetobacter baumannii, and Pseudomonas aeruginosa across matched wild-type, hyperporinated, efflux-deficient, and combined strain backgrounds, so that outer-membrane entry and efflux contributions can be resolved separately under a common protocol. Baseline models predicting accumulation from chemical structure, conditioned on species and barrier state, would then be improved through active-learning rounds that select the most informative compounds for testing, with cross-species transfer reported explicitly and the resulting data, protocols, genotype-resolved strain metadata, and models released as an open community benchmark. This proposal is a product from “Roadmapping the Future of AI-Driven Antimicrobial Resistance Research”, a community alignment initiative led by The Align Foundation and supported by Google DeepMind, with the goal of prioritizing key challenges and opportunities for the use of AI/ML in helping combat AMR.
Language: English
Subfield: Molecular Medicine
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7202185309