#12 · ML / AI methods

Machine learning links resistance phenotypes and genotypes in bacterial AMR

The dominant thread is machine learning applied to bacterial resistance phenotypes and genotypes. CarbaDetector, a random-forest model, predicts carbapenemase production in Enterobacterales from disk diffusion inhibition zones and is reported to outperform EUCAST interpretation, working from a routine phenotypic readout. AMR-MoEGA couples a mixture-of-experts resistance predictor with a genetic algorithm that simulates resistance evolution across genomes, placing evolutionary search alongside prediction. That coupling of simulated evolution with predictive architecture is the new element here. The tuberculosis study is observational and epidemiological, examining clinical and bacterial determinants of unfavorable treatment outcomes in Georgia, and does not share the modeling thread. The listed items do not report interpretability analyses, so claims about model transparency remain untested.