Branislava Jankovic*, Aleksandr Matsun, Noël Malod-Dognin and Natasa Przulj
Mohamed Bin Zayed University of Artificial Intelligence
branislava.jankovic [at] mbzuai.ac.ae
Abstract
Pseudomonas aeruginosa is one of the most dangerous antibiotic-resistant bacteria in the world, classified by the WHO as a critical priority pathogen. It causes severe and often fatal infections in hospitalised and immunocompromised patients, and it is notoriously difficult to treat. Its resistance arises from multiple mechanisms working simultaneously — a near-impermeable outer membrane that blocks drugs from entering, powerful efflux pumps that actively expel antibiotics before they can act, and the capacity to rapidly acquire new resistance genes — leaving clinicians with very few therapeutic options.
To address this at a systems level, we applied Non-negative Matrix Tri-Factorization (NMTF), a machine learning method that integrates multiple biological datasets simultaneously, including drug chemical similarities, drug-target interactions, pathogen resistance profiles, point mutations, and protein co-expression networks. By learning hidden patterns across all data types at once, NMTF predicts novel drug-target associations and identifies bacterial vulnerabilities that are not visible in any single dataset alone, giving us a data-driven map of resistance mechanisms and potential new targets. Guided by these predictions, we selected five targets in P. aeruginosa covering both resistance and essential survival processes: MexA and MexB, the components of the MexAB-OprM efflux pump that actively pumps antibiotics out of the cell; MurA, which catalyses the first committed step in peptidoglycan biosynthesis and is essential for cell wall integrity; PonA, a penicillin-binding protein responsible for cell wall cross-linking; and LpxA, which initiates lipid A biosynthesis and is indispensable for outer membrane assembly.
For each target, new drug candidates were generated and evaluated through virtual screening, molecular docking, and ADMET profiling. Experimental validation of these compounds holds considerable therapeutic potential — efflux pump inhibitors targeting MexA and MexB would restore the efficacy of existing antibiotics by disrupting the bacterium’s primary resistance mechanism, while compounds directed against MurA, PonA, and LpxA would compromise essential biosynthetic processes indispensable for bacterial viability.Together, this multi-target strategy reduces the likelihood of resistance developing to any single drug, and offers a genuine computational path toward treating infections that are currently untreatable.
Keywords: Pseudomonas aeruginosa, NMTF, antibiotic resistance