mmseqs2

MMseqs2 clusters sequences fast enough to be run over a whole benchmark. That is why it is in this corpus: honest evaluation of a structure-QA model requires splits held out by sequence similarity rather than at random, and an unclustered split quietly inflates every number downstream of it.

One Bioconda package, locked green, and a BioContainer pulled to check it is there — the digest is in the frontmatter.