Skin microbiome catalog fills a reference-genome gap: 452,000 genomes, most species previously unrepresented

The skin microbiome now has a multi-kingdom reference genome catalog spanning bacteria, archaea, fungi and viruses across dry, moist and sebaceous skin microenvironments and 10 dermatological conditions. Earlier skin-specific catalogs such as SMGC, UHSG and ELSG improved read-mapping rates but remained limited in non-bacterial domains, global strain-level variation and niche-specific adaptation. The catalog was built from 12,409 newly sequenced samples plus 4,396 public skin metagenomes and 3,077 isolate genomes, reconstructing 452,397 metagenome-assembled genomes, of which 253,565 prokaryotic genomes clustered into 11,891 prokaryotic species-level genome bins.

Many skin microbes with no prior reference genome are now represented, including 7,124 prokaryotic species-level genome bins with no match in SPIRE or GTDB. These uncharacterized bins account for 59.91% of the 11,891 prokaryotic species-level genome bins, carry 51,427 genomes, and increase total phylogenetic tree branch length by 76.9%. The uncharacterized bins appear at all seven dry sites and multiple moist and sebaceous sites, and average 18.2% of recovered genomes across the 10 surveyed skin diseases.

Functional dark matter that sequence similarity search cannot match is turned into structural candidates for inferring what these proteins may do. Traditional sequence-alignment methods often fail on microbial proteins with no homology to existing databases, leaving this functional repertoire largely unexplored. Of 1,272,886 protein families, 79,434 had no hit in Pfam-A or RefSeq; ESMFold modeling yielded 1,018 structures with mean pLDDT above 90, and AlphaFold3 re-evaluation of 74 proteins with high-quality pockets confirmed 30 structures.

The resource targets skin microbiome and dermatology research, covering dry, moist and sebaceous microenvironments and 10 skin disease states, with Chinese cohorts contributing the largest share. Metagenome-assembled genomes are computational inferences, and the authors note possible fragmentation or mis-binning, particularly for mobile genetic elements and accessory genomes. Functional annotations and small-molecule binding predictions are in silico hypotheses that require biochemical characterization and in vivo models.

The authors state that although sampling spans diverse geographic and clinical conditions, the heterogeneity of the global human population means the true boundaries of cutaneous microbial diversity remain to be fully charted. Viral genomes are currently available at a public link and will be deposited in a permanent public repository before publication. Links between uncharacterized lineages and disease states are described only as partly associated with distinct disease contexts, not as causal.