#23 · ML / AI methods

Inferring microbial taxonomy directly from long-read and metagenomic sequencing data

The shared thread is inferring microbial taxonomic identity directly from sequencing reads, either by referencing curated databases or by applying learned sequence models. HiFiTaxa is a Nextflow pipeline for PacBio HiFi full-length 16S rRNA and ITS amplicons, referenced to GTDB and UNITE, and benchmarked against BLCA, naive Bayes, Emu and EMITS. Taxonova is a deep learning classifier for species-level amplicon taxonomy across short- and long-read lengths, evaluated on mock communities. MicrobeRT fine-tunes Nucleotide Transformer, DNABERT and HyenaDNA for metagenomic read taxonomy and resistance gene detection. What is new is the pairing of long-read full-length amplicons with pretrained genomic language models and explicit comparison against established classifiers. The three studies differ in target, method and evidence, so this reflects a shared direction rather than a common algorithm.