2bLCA and top hit (--lca-mode) differ in search sensitivity

[apcamargo]

I'm comparing MMSeqs2 taxonomic assignment with approx. 2bLCA and top hit and noticed that the later approach classifies more genes than the former. I extracted the alignments using --extract-lines 1 and the top hit search had more hits to the database. All parameters were the same with the exception of --lca-mode.

Example:

mmseqs taxonomy querydb/querydb gtdb_r202/gtdb_r202 taxonomydb/taxonomydb tmp -s 3.0 --lca-mode 3 --tax-output-mode 2 --threads 64

Is behavior expected? If so, what is causing this difference?

I'm using release 13-45111.

Thanks!