What it is
MMseqs2 is the canonical name of a sequence-search and clustering method used in the PaperBLAST Data Explorer — Literature Coverage Bias in Protein Sequence Space analysis; no additional alias or stable external identifier is reported in that document. 1
Key facts in PaperBLAST analysis
MMseqs2 clustered 815,571 protein sequences from the kescience-paperblast collection at 90%, 50%, and 30% sequence-identity thresholds. 1
The clustering produced 628K clusters at 90% identity, 345K clusters at 50% identity, and 215K clusters at 30% identity. 1
The report interprets these thresholds as representing low strain-level redundancy at 90% identity, a protein-family scale at 50% identity, and a superfamily scale at 30% identity; the 50% cutoff is conventional rather than a universal biological boundary. 1
At 90% identity, 83.3% of clusters were singletons; at 50% identity, 67.5% were singletons; and at 30% identity, 62.8% were singletons. 1
The largest 90%-identity cluster contained 212 members and was identified as actin. 1
At 50% identity, the largest reported protein families included HSP70/BiP with 650 members, GAPDH with 609, enolase with 552, and GroEL with 443. 1
The 50%-identity clustering supported the identification of poorly studied protein families relevant to gene-function-acquisition-depth: 9.2% of families (31,653) had zero papers across all members, 46.1% (159,046) had exactly one paper, and 4.3% (14,904) had 20 or more papers. 1
The analysis found 5,218 multi-member 50%-identity families, representing 14,534 sequences, with no literature whatsoever. 1
Family size was positively associated with literature coverage: 95.4% of multi-member 50%-identity clusters had at least one member with a paper, while 4.6%—5,218 families—had none. 1
The report describes the dark families as dominated by REBASE methyltransferases and biolip structural entries, while cautioning that the 50% identity threshold is an analytical convention and that missing literature may reflect incomplete text mining rather than genuine absence of study. 1
Prophage ecology application
The Prophage Gene Modules and Terminase-Defined Lineages Across Bacterial Phylogeny and Environmental Gradients analysis extends MMseqs2’s demonstrated protein-family use to prophage lineage analysis: it clustered 38,085 TerL sequences from 11,789 species into 10,991 lineages at 70% AAI, where AAI means amino-acid identity. 2
Threshold sensitivity produced 4,001 lineages at 50% AAI and 16,283 at 80% AAI; the largest 70%-AAI lineage contained 1,094 members across 869 species, while 6,921 lineages (63%) were singletons. 2
This application refines the existing interpretation of identity thresholds: MMseqs2 supported both broad protein-family characterization and finer TerL lineage partitioning, but the prophage analysis found no significant environment-specific enrichment for individual TerL lineages in 0/500 tests after FDR correction. 2
Relation to the source resource
MMseqs2 provided the sequence-family layer for the collection-level characterization of kescience-paperblast, allowing literature coverage to be evaluated across protein families rather than only individual genes. 1
In the prophage study, it provided the lineage-clustering layer for comparing TerL-associated prophage variation across hosts and environments, complementing module-level ecological analyses. 2