Summary
A bioRxiv preprint presents Sma3s v3, a scalable workflow for assigning functions to proteins in proteomes, pangenomes and metagenomes. In tests on a Vibrio cholerae pangenome and a metagenomic catalogue, it reported higher annotation coverage than the cited comparator tools.
Researchers have presented Sma3s v3, a computational workflow for assigning biological functions to proteins in large sequence collections. In a bioRxiv preprint posted on September 19, 2026, the authors report that the method annotated 60.8% of gene clusters in a large Vibrio cholerae pangenome and 86.3% of proteins in a metagenomic catalogue, exceeding the coverage reported for the comparator tools used in the study.
Functional annotation links a protein sequence to likely roles, such as participation in a biochemical process or interaction with a cellular structure. The task is increasingly difficult as sequencing produces datasets containing millions of predicted proteins, many of which have no experimentally characterised counterpart.
A multi-stage workflow for large sequence collections
Sma3s v3 is a reimplementation of the Sma3s annotation strategy. It combines three sources of evidence: transfer from highly similar homologous proteins, inference based on orthologous proteins, and functional enrichment among groups of homologues. Homologues are proteins related through evolution, while orthologues are corresponding genes in different organisms that can provide evidence about conserved function.
The new version replaces BLAST-based searches with MMseqs2 and adds parallel processing, reusable SQLite caches and taxonomic filtering. Its outputs retain the evidence behind each assignment, allowing an annotation to be traced back to the sequence relationships or other information used to generate it.
A pangenome represents the combined gene repertoire found across multiple genomes of a species or group. A metagenome is a mixed collection of genetic material from an environmental or host-associated community. Both can contain a large core of familiar genes alongside accessory or unique sequences that are harder to annotate.
Results from a bacterial pangenome and metagenome
For the pangenome test, the researchers used 50,415 gene clusters from 11,295 quality-filtered Vibrio cholerae genomes. After non-informative assignments were excluded, Sma3s v3 annotated 30,662 clusters, or 60.8%. InterProScan annotated 60.2% under the comparison, while eggNOG-mapper annotated 41.4%. The Sma3s workflow also produced 5,747 annotations that neither comparator recovered.
Comparisons using Gene Ontology terms showed broad semantic agreement among the methods. The authors report that Sma3s v3 often contributed more non-redundant information in the Molecular Function and Biological Process categories. Within the pangenome, coverage reached 97.1% for core clusters and approximately 59% for accessory and unique clusters.
The analysis also identified 1,838 clusters with exact protein matches in non-Vibrio genera. These were treated as candidate horizontally transferred clusters and were enriched for functions associated with genetic mobility, antimicrobial resistance and metal tolerance. The result provides a way to prioritise unusual sequence patterns for further biological investigation; it is not, by itself, a direct experimental demonstration of transfer.
The second test used a metagenomic catalogue containing 843,935 proteins. Sma3s v3 annotated 728,014 proteins, or 86.3%, compared with 616,895 proteins, or 73.1%, for InterProScan 2026. The authors also report that the workflow recovered approximately 20,000 functional terms not found in that comparison.
The study presents these results as evidence that Sma3s v3 can support large-scale annotation and re-annotation while retaining a record of the evidence behind individual assignments. The evaluation is reported in a bioRxiv preprint and covers the specified V. cholerae pangenome and metagenomic catalogue; performance in other datasets will depend on their sequence diversity and the annotation criteria applied.