Milot Mirdita

If you use conda, please make sure to also use the conda-compilers and not your system compiler. before calling `cmake` please do: ``` export CC=path-to-conda-gcc export CXX=path-to-conda-g++ ```

Can you try to use less threads (`--threads 32` or `64`) on the same machine?

You can use `--max-accept` for this (number of hits allowed to pass through the alignment stage). One more parameter that is useful here is `--max-rejected`, the number of rejected hits...

`--sort-results` is linclust parameter that is confusingly visible within the `search` workflow, please ignore that. And regarding `--max-accept`, i understand now what you want to achieve, sorry it doesn't quite...

How large is the database you created? Would it be possible to share? How does your tax mapping look like (`UVIG_taxid_mapping_cleaned`). It seems to create some very large taxid values...

This looks about right. MMseqs2 creates one output file per thread and doesn't bother merging them in the end if its not needed. You can use `easy-cluster` instead of `cluster`...

They are computed by the AFLP/FALP library (https://academic.oup.com/bioinformatics/article/32/2/304/1744607). The computation takes a bit for new substitution matrices, so we precompute them for commonly used matrices and hardcode the values.

I fixed the taxonomy issue, we were using an unsigned short for the source file names. This changes was introduced a while ago and I had forgotten that we use...

The issue here is likely that your sequences are very short and our prefilter struggles with these. You can try to force 6-mers with a short spaced k-mer pattern (see...

Does this issue also happen in release 14?