Rapid RNA Gene Panel Methodology to Aid Antimicrobial Stewardships

Using a genetic algorithm for feature selection plus automated machine learning, the authors distilled the full transcriptome (~6,000 genes) down to antibiotic-specific panels of ~35–40 genes, after which performance plateaued. On held-out test sets, accuracy was ~99% for meropenem and ciprofloxacin and ~96% for tobramycin and ceftazidime, improving on full-transcriptome baselines (accuracy up to 0.90; F1 up to 0.88). Only 2–10% of selected genes overlapped the Comprehensive Antibiotic Resistance Database; panels included canonical markers (e.g., ampC, mexAB-oprM) alongside novel biomarkers (e.g., gbuA, fpvA), and mapped to regulatory modules linked to efflux, DNA repair, oxidative stress, ribosomal function, and metabolic adaptation (79–88% coverage in PA14 iModulons). For pharmacy operations, the key translational insight is that a targeted expression assay covering ~40 genes per antibiotic could be more practical than whole-transcriptome profiling and, if paired with rapid RNA workflows, might shorten time to susceptibility guidance to support earlier optimization/de-escalation. Caveats for informatics and stewardship leaders: this is an internal validation on a public dataset (no external or prospective testing), it’s a preprint, there is no turnaround-time or cost analysis, and biomarkers may reflect correlated physiology rather than causal resistance—so implementation would require external validation, analytical/clinical verification, and careful integration with antimicrobial stewardship decision support.

Shahreen N, Shahid SA, Subhani M, Al-Siyabi A, Saha R. Minimal Gene Signatures Enable High-Accuracy Prediction of Antibiotic Resistance in Pseudomonas aeruginosa. Preprint. bioRxiv. 2025;2025.04.29.651273. Published 2025 May 3. doi:10.1101/2025.04.29.651273