Abstract

The rise of antimicrobial resistance (AMR) demands predictive frameworks that can guide the rational design of antibiotics against Gram-negative pathogens. A major challenge lies in disentangling the interplay between permeability through porins, intracellular accumulation, and antibacterial activity. Here, we present an interpretable artificial intelligence (AI) approach that integrates experimental and computational data to build predictive models of permeability and accumulation in Escherichia coli. By unifying these models, we uncover systematic relationships between permeability and activity (MIC), accumulation and activity, as well as accumulation and permeability. Our interpretable machine learning framework emphasizes transparent feature selection and descriptor-based reasoning, enabling mechanistic insights rather than black-box predictions. This strategy provides a scalable path toward identifying molecular determinants of antibiotic efficacy, ultimately accelerating the discovery of compounds with improved accumulation and activity profiles against Gram-negative bacteria.