Comments on: HMMER 3.0b3: the final beta test release

By: Sean Eddy

Sean Eddy — Thu, 12 Aug 2010 11:54:36 +0000

You’d need to give more information, enough that we could reproduce your problem, including what version of the code you have, and what machine you’re on.

By: Eshita

Eshita — Thu, 12 Aug 2010 06:46:45 +0000

I have created a profile HMM database and hmmscan option (HMMER3) is working fine on it , except when I want to use the –domtblout option. But this option is working with Pfam HMM db. I made the HMM db using hmmbuild, hmmpress.

It would be great if you can give some insight on this.

By: Eshita

Eshita — Mon, 08 Feb 2010 08:09:59 +0000

Thanks a lot ,Eddy

By: Sean Eddy

Sean Eddy — Fri, 05 Feb 2010 17:17:36 +0000

There’s two kinds of thresholds: reporting and inclusion. The reporting threshold controls what appears in the default output. The inclusion threshold controls what domains get marked as “significant” in domain output (! instead of ?), what appears in optional output alignments (-O) from search programs, and what does into the next round of jackhmmer. The reporting threshold would generally be less stringent than the inclusion threshold.
In the per-sequence output, where you’re looking, the inclusion threshold gets marked; this is the line where HMMER thinks you should stop looking at the hits. Everything above it should be real (but isn’t necessarily, of course), and everything below it should be looked at with a jaundiced eye.

By: Eshita

Eshita — Thu, 04 Feb 2010 12:08:48 +0000

Hi,

I ran hmmscan (HMMER3.0b3) with an E-value of 0.1 but got some results mentioning “inclusion threshold”. Can you please clarify what is that for. An example of it appended below.

*********************
Scores for complete sequence (score includes all domains):
— full sequence — — best 1 domain — -#dom-
E-value score bias E-value score bias exp N Model Description
——- —— —– ——- —— —– —- — ——– ———–
—— inclusion threshold ——
0.012 12.2 0.0 0.014 12.0 0.0 1.1 1 51182
**********************

Thanks

By: Sean Eddy

Sean Eddy — Wed, 03 Feb 2010 01:03:19 +0000

Oi vey. OK, that’s an important clue. That puts the blame on GNU autoconf, and means we must have a system in our compile farm that behaves the same way yours does. We’ll look into it again, thanks.

By: Michael Kuhn

Michael Kuhn — Tue, 02 Feb 2010 09:49:17 +0000

Sorry, I didn’t state that I compiled on x86_64 in the first place, and _FILE_OFFSET_BITS was not set. I just ran configure again without command line options, and this is part of the output:

checking for _FILE_OFFSET_BITS value needed for large files… no

So also on my system the test configure uses doesn’t reflect reality.

By: Sean Eddy

Sean Eddy — Sat, 30 Jan 2010 14:27:45 +0000

Michael – thanks, we’re puzzled what’s happened there. The ./configure sets _FILE_OFFSET_BITS automatically. We’ve verified that it configures properly if you rebuild from source. Somehow the x86_64 precompiled binaries for 3.0b3 don’t have the flag set properly, and we’re not sure how that happened. We’ll keep an eye out for it in the 3.0 build.

Hiroshi – the domtblout format has the query and target lengths; so use –domtblout to get what you want here. In the “human readable” standard output, line length is at a premium, and we don’t have room, hence the [], .], [., .. code.

By: Hiroshi

Hiroshi — Wed, 27 Jan 2010 18:17:45 +0000

I just thought it’s nicer to show the length of the domains (the information found in the .hmm files) in the hmmscan output. By comparing it with the span of the hit on sequences, one can judge if the domain is almost full-length or truncated. Although there is a field “[]” “[." ".]” “..”, it does not tell how much amino acid left in the termini.

By: Michael Kuhn

Michael Kuhn — Tue, 26 Jan 2010 14:42:04 +0000

When trying to run hmmpress on a large concatenated .hmm file (3 GB, still have to figure out if this a good idea), I got this error:

Fatal exception (source file ../../src/p7_hmmfile.c, line 782):
ftello() failed

I solved it by passing CFLAGS=’-D _FILE_OFFSET_BITS=64′ to ./configure and recompiling, so it looks like seeking beyond 2 GB doesn’t work by default. Perhaps you could set this by default if HMMER is being built on a 64-bit architecture.