Nhmmer: different (2x smaller) e-values with fasta and hmmerdb

[Augustin-Zidek]

Nhmmer reports different e-values (consistently off by an exact factor of 2) when I run against a fasta and a hmmerdb made from that fasta using makehmmerdb.

Reproduction

1. Download rfam, let's call it rfam.fasta.

2. Produce HMMERDB from the downloaded fasta:

   makehmmerdb rfam.fasta rfam.hmmerdb

3. Run Nhmmer against the hmmerdb:

   nhmmer --tblout /tmp/tblout.txt --rna --watson -A /tmp/output.a3m /tmp/query.a3m rfam.hmmerdb

4. Run Nhmmer against the fasta:

   nhmmer --tblout /tmp/tblout.txt --rna --watson -A /tmp/output.a3m /tmp/query.a3m rfam.fasta

Expected result Since I ran with --watson, I would expect the e-values to match, but they are consistently off by a factor of 2, with the hmmerdb being 2x larger.

More observations

• There are 12 cases: {watson, crick, both} × {fasta, hmmerdb} × {z-value set, z-value unset}
• If I set the z-value, the e-values are consistent.
• If I don't set --watson or --crick, the e-values are also consistent.
• I would expect the e-values to not depend on whether I used fasta or hmmerdb.
• I suspect the bug might be around here: https://github.com/EddyRivasLab/hmmer/blob/master/src/nhmmer.c#L1108

[cryptogenomicon]

Thanks (and thanks for the concise and excellent summary). I've fixed that in our develop branch, and the fix will appear in the next release.

[Augustin-Zidek]

Thanks for the quick fix!