TaxDistill: Improving Metagenomic Taxonomic Annotation via Distilled Genomic Foundation Models

Rongye Ye, Lun Li, Zheng Luo, Yiran Zhan, Shuhui Song, These authors contributed equally to this work · 2026

Read TaxDistill: Improving Metagenomic Taxonomic Annotation via Distilled Genomic Foundation Models by Rongye Ye, Lun Li, Zheng Luo, Yiran Zhan, Shuhui Song, These authors contributed equally to this work (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-28868.txt.

Description

Rongye Ye and five colleagues present TaxDistill, a plug-and-play teaching framework for microbe taxonomic annotation. Match tools are fast but noisy, and fix methods trained on hard pseudo-labels inherit that noise. TaxDistill uses the GenomeOcean model as a meaning teacher, updating by layered rules while its soft odds guide a light student network. The design curbs student overfit and supports cautious rejection of shaky calls. On seven CAMI2 datasets TaxDistill improves trust over baselines, lifting MMseqs2's F1 score on the Gastrointestinal dataset to 0.941.

Catalog record

CREATORRongye Ye, Lun Li, Zheng Luo, Yiran Zhan, Shuhui Song, These authors contributed equally to this work
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE46436
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.28868

Browse the full library · About the Tower of Babel