[FEATURE REQUEST] A command to make new HMM sources from a list of COG IDs. about anvio

[meren]

It seems to me that it would be suffucient to have a wrapper for hmmbuild and some downstream code to turn the resulting HMM(s) to conform with user-defined hmm-source file structure. A script like anvi-script-pfam-accessions-to-hmms-directory but anvi-script-fasta-to-hmms-directory.

[meren]

So essentially with this kind of feature anvi'o would be able to take a bunch of gene sequences from anywhere, align them, hmmbuild, and turn the outuput in an hmm-source, am I understanding it correctly?

Much like the (I think) defunct anvi-script-pfam-accessions-to-hmms-directory

I didn't know it was defunct? I just used it a few days ago 😂

[Ge0rges]

Apologies, I think I' either thinking of a different script or had trouble accessing Pfam and assumed the merge broke the script.

Yes, you're understanding is correct.

[dspeth]

Came here for a different issue, but saw this and wanted to drop my two cents. At least for some COGs, the definition is quite broad (see the somewhat frequent "or related enzyme" appendix to the function name) and that might cause problems when these are turned into HMMs without further curation.
if a feature like this is (primarily) user facing, that might cause some data analysis issues

[meren]

Thanks for your input, Daan. That's also my concern since the sensitivity and specificity of the models will be all over the place since the number of genes and their conservancy within COGs are all over the place. I wonder what is it we could do to make sure people don't reach misleading conclusions with a tool like this if implemented.

[dspeth]

Agreed. I was thinking along the lines of automated sequence dataset cleanup and wonder if there could be sanity checks applied to the genes retrieved, to see whether it is sensible that they indeed represent homologous proteins. I have no immediate good suggestions, other than a blunt length assessment.

I tried average pairwise identity the other day, but that wasn't very helpful for the use case I was looking at. All v All pairwise alignment does help identify outlier sequences as those only aligning to a small subset of the tested sequences.

[Ge0rges]

What about clustering with an identity threshold akin to usearch cluster_fast?

[Ge0rges]

Perhaps a reduced scope would be a script that makes an HMM given protein sequences, and their corresponding gene names, at least to begin with.