Fig. 1: A metabolic systems approach for genetic associations.
From: A biochemically-interpretable machine learning classifier for microbial GWAS
a In this study, data describing TB genome sequences and AMR data types are integrated with a metabolic model to learn a biochemically-interpretable classifier, named Metabolic Allele Classifier (MAC). The MAC parameters consist of allele-specific flux capacity constraints, a, and an antibiotic-specific metabolic objective, c, both of which are inferred from the data. b The optimal MAC describes strain-specific polytopes in flux space that separate into resistant (R) and susceptible (S) regions. The MAC objective function, cTv, is identified as normal to the plane that best separates R and S. c The learned MAC provides a biochemically-based hypothesis of AMR mechanisms and allele-specific effects through interpretation of c and v. The genome-scale flux state of a strain, v, consists of fluxes that are directly activated by alleles (allelic fluxes) and those that are flux-balance consequences of the allele-activated fluxes (compensatory fluxes). Abbreviations: S, susceptible; R, resistant; AMR, antimicrobial resistance.
