TaxDistill: Improving Metagenomic Taxonomic Annotation via Distilled Genomic Foundation Models

2026-05-22

AI Summary

This study addresses the limitations of current metagenomic classification methods, which suffer from high microbial diversity, incomplete reference databases, and noisy labels, leading to degraded representation learning and classification performance. To overcome these challenges, the authors propose TaxDistill, a novel framework that leverages GenomeOcean—a large-scale genomic foundation model with 500 million parameters—as a teacher network to generate soft labels via knowledge distillation and transfer them to a lightweight student network, thereby mitigating label noise introduced by initial retrieval tools. Extensive experiments on seven CAMI2 datasets demonstrate that TaxDistill significantly outperforms existing approaches; for instance, it improves the F1 score of MMseqs2 from 0.763 to 0.941 on gastrointestinal samples, surpassing baselines such as Taxometer and substantially enhancing species annotation accuracy in complex environmental samples.

Abstract

Metagenomic taxonomic annotation aims to identify the microbial origins of DNA fragments in environmental samples. Traditional methods that rely on sequence similarity are often constrained by the high microbial diversity and the incompleteness of reference databases, which has motivated the development of learning approaches such as Taxometer that perform post hoc correction to learn more informative metagenomic sequence representations. However, these methods typically rely on labels derived from similarity search tools during training, which inevitably introduces noise that can impair representation learning and degrade classification performance. To address this issue, we propose TaxDistill, a knowledge distillation framework for metagenomic classification. We introduce GenomeOcean, a 500M parameter genomic foundation model, as the teacher network to extract deep semantic features and generate soft labels based on confidence. By distilling this soft label information into a lightweight student network, TaxDistill effectively reduces the label noise introduced by initial retrieval tools. Comprehensive experiments on seven diverse CAMI2 datasets demonstrate that TaxDistill outperforms existing baselines in most scenarios. For instance, on the Gastrointestinal dataset, it improves the F1 score of MMseqs2 from 0.763 to 0.941, outperforming the Taxometer baseline. Overall, TaxDistill provides a reliable method for label correction in complex metagenomic analysis.

Problem

• metagenomic taxonomic annotation
• label noise
• sequence similarity
• reference database incompleteness
• microbial diversity

Innovation

• knowledge distillation
• genomic foundation model
• metagenomic classification
• soft labels
• label noise reduction