Seeking Correlation Among Porin Permeabilities and Minimum Inhibitory Concentrations Through Machine Learning: A Promising Route to the Essential Molecular Descriptors

Developing effective antibiotics against Gram-negative bacteria remains challenging due to their protective outer membrane. With this study, we investigated the relationship between antibiotic permeation through the OmpF porin of <i>Escherichia coli</i> and antimicrobial efficacy. We measured the relative permeability coefficients (RPCs) through the bacterial porin by liposome swelling assays, including non-antibacterial molecules, and the minimum inhibitory concentrations (MICs) against <i>E. coli</i>. We developed a machine learning (ML) approach by combining classification and regression models to correlate these data sets. Our strategy allowed us to quantify the negative correlation between RPC and MIC values, clearly indicating that increased permeability through OmpF generally leads to improved antimicrobial activity. Moreover, the correlation was remarkable only for compounds with significant permeability coefficients. Conversely, when permeation ability is low, other factors play the most significant role in antimicrobial potency. Importantly, the proposed ML-based approach was set by exploiting the available seminal information from previous investigations in order to keep the number of molecular descriptors to the minimum for greater interpretability. This provided valuable insights into the complex interplay between different molecular properties in defining the overall outer membrane permeation and, consequently, the antimicrobial efficacy. From a practical perspective, the presented approach does not aim at identifying the "golden rule" for boosting antibiotic potency. The automated protocol presented here could be used to inspect, in silico, many alternatives of a given molecular structure, with the output being the list of the best candidates to be then synthesized and tested. This could be a valuable in silico tool for researchers in both academia and industry to rapidly evaluate novel potential compounds and reduce costs and time during the early drug discovery stage.