一种融入分类学信息的稀疏 DNA 基础模型，用于微生物基因组学
A Taxonomy-Informed Sparse DNA Foundation Model for Microbial Genomics
微生物对生态系统和人类健康至关重要，但其种类极多、各类群在训练数据中分布极不均衡，给 DNA 序列的机器学习建模带来困难。作者提出 MicroGlot，一个融入微生物分类学信息的 DNA 基础模型，基于 370 万条序列、约 3783 亿个核苷酸、覆盖 99700 个物种预训练。它用双曲嵌入编码类群之间的层级关系，并采用稀疏“专家混合”架构。零样本评估显示，模型各层表示能编码表型特征和分类身份；与去掉分类学信息的对照模型相比，加入分类学知识持续提升表示质量。在 1000 个物种上的路由指纹也比四核苷酸组成更贴合分类群。该研究为微生物基因组分析提供了高效的基础模型，但尚未经同行评审。
为什么推荐给您：构建融入分类学信息的微生物 DNA 基础模型，属重要数据资源与建模进展，尚未经同行评审。
不需要生物学背景，多打比方
正在获取全文并生成讲解（拿不到全文就依据摘要），大约需要 30–60 秒…
已等待 0 秒
这篇还没有动画
动画会把研究的流程、作用机制和关键结果一步一步演示出来，每一步都标明出自原文哪里。制作大约需要 30–60 秒。
摘要Abstract
Microorganisms are indispensable to terrestrial ecosystems, with their genomic material underpinning critical functions and applications across agriculture, biotechnology, and human health. Although genomic language models have advanced representation learning in DNA sequences, the extensive diversity of microorganisms and imbalanced taxonomic representation in the pretraining corpora pose challenges for effective microbial genomic sequence modeling. Here we present MicroGlot, a taxonomy-informed microbial DNA foundation model pretrained on 3.70 million sequences comprising 378.3 billion nucleotides across 99{,}700 species. MicroGlot encodes the hierarchical relations among taxa through hyperbolic embeddings, incorporating microbial taxonomic knowledge into a sparse mixture-of-experts architecture. Zero-shot evaluation of MicroGlot's layer embeddings demonstrates that the model's representations encode phenotypic traits and taxonomic identity. Comparison with a taxonomy-ablated variant trained under the same pretraining scheme shows that incorporating taxonomic knowledge consistently improves representation quality across the layers of MicroGlot. MicroGlot also combines optimized training techniques with efficient architectural components from modern large language models, achieving leading zero-shot performance across layers and competitive fine-tuning performance with low computational overhead. In a 1000-species set sampled from major cellular domains and viral realms, MicroGlot's routing fingerprints show greater agreement with taxonomic groups than tetranucleotide composition, reflecting taxonomically structured expert routing in multilingual modeling of microbial genomes. Overall, we show that MicroGlot serves as an efficient and effective DNA foundation model for microbial genomic analysis.