beta
/K-mer-based models show marginally higher classification accuracy (0.854–0.855 vs 0.828 for ASVs)