Human Gut Microbial Protein Structure Database (GMPS) part1

The human gut microbiome contains numerous proteins whose functions remain elusive yet are pivotal for host health. Sequence-based methods often falter when attempting to infer functions within this microbial proteome due to evolutionary divergence. To address this challenge, we develop the Human Gut Microbial Protein Structure Database, which incorporates ~2.7 million predicted protein structures. Our findings reveal that structural analogy enhances the annotation of phage proteins. We detail the structural diversification of phage endolysins and confirm their potential in eliminating gut pathobionts. Furthermore, our structure-guided approach is effective in the identification of microbial-host isozymes. By employing structural alignments, we identify previously unrecognized bacterial enzymes involved in melatonin biosynthesis. Finally, we present an alignment-free method, Dense Enzyme Retrieval, based on structure-encoded protein language models for ultrafast and sensitive detection of remote homologs. Our research underscores the value of computational structural genomics in elucidating the functional landscape of the human gut microbiome.人类肠道微生物组含有大量功能尚未明确，但对宿主健康至关重要的蛋白质。由于进化分化的影响，基于序列的方法在推断该微生物蛋白质组的功能时往往效果不佳。为应对这一挑战，我们构建了人类肠道微生物蛋白质结构数据库（Human Gut Microbial Protein Structure Database），该库收录了约270万个预测得到的蛋白质结构。研究结果表明，结构类比可提升噬菌体蛋白质的功能注释效率。我们详细解析了噬菌体内溶素的结构多样性，并证实其在清除肠道致病共生菌方面具有应用潜力。此外，我们基于结构的研究方法可有效识别微生物-宿主同工酶。通过结构比对，我们发现了此前未被报道的、参与褪黑素生物合成的细菌酶类。最后，我们提出了一种基于结构编码蛋白质语言模型（structure-encoded protein language models）的无比对方法——Dense Enzyme Retrieval，可实现超快速且高灵敏的远程同源物（remote homologs）检测。本研究凸显了计算结构基因组学在解析人类肠道微生物组功能图谱中的重要价值。