A new way to identify microbes with incomplete data

HyperTaxa uses multiscale genome representations to help identify microbes even when reference coverage is incomplete.

Reliable microbial taxonomic inference is challenged by the rapid growth of newly sequenced organisms. Incomplete reference coverage also makes identification difficult.

HyperTaxa is a taxonomy-aligned representation learning framework for open-set microbial taxonomic inference. It encodes whole genomes as multiscale hyperspectral k-mer transition fields.

The framework aligns these representations to a taxonomy-derived semantic space. This uses a two-stage Node2Vec-to-MDS procedure. This procedure combines local graph structure with global hierarchical distance preservation.

HyperTaxa supports zero-shot inference over species excluded from genome-level training. It also supports generalized zero-shot inference and cascaded genus-to-species prediction.

The framework improves taxonomic prediction over pseudo-image and transformer-graph baselines. The strongest gains occur at higher taxonomic ranks.

Multiscale transition encoding and taxonomy-aware alignment are key contributors to generalization. HyperTaxa can serve as a candidate-generation framework under incomplete reference coverage and open-set uncertainty.

Key facts

  • HyperTaxa supports zero-shot inference over species excluded from genome-level training.
  • The framework improves taxonomic prediction over pseudo-image and transformer-graph baselines.
  • Multiscale transition encoding and taxonomy-aware alignment contribute to generalization.

All sections

Edition