Figure 1: A typical machine learning workflow for AMR phenotype detection.
From: Antimicrobial Resistance Prediction in PATRIC and RAST
Genomes for a given species are binned according to whether they are resistant or susceptible to an antibiotic and the k-mer counts are computed for each genome. The k-mer counts are then merged to form a matrix. A machine learning algorithm searches this matrix to find the k-mers that distinguish the resistant and susceptible genomes. These distinguishing k-mers are then used as a “classifier” to predict the phenotype for a new genome.
