SEA-PHAGES | Deep TMHMM?

Link to this post \| posted 25 Apr, 2022 19:04
plconnerly	The TMHMM 2.0 page linked from the Bioinformatics Guide contains a note at the top stating the TMHMM 2.0 is outdated and giving a link to Deep TMHMM (https://dtu.biolib.com/DeepTMHMM). Is either version ok for searching for transmembrane domains for calling membrane functions, or should we stick to TMHMM 2.0 because it is specified in the guide? Thanks!

Link to this post \| posted 25 Apr, 2022 19:06
debbie	Either and all work. debbie

Link to this post \| posted 25 Apr, 2022 19:35
plconnerly	Thanks, Debbie! I went ahead and tried both on a small set of proteins and found an interesting difference. This preprint https://doi.org/10.1101/2022.04.08.487609 about Deep TMHMM mentions issues with signal sequences being erroneously called as transmembrane domains. Although TMHMM 2.0 and Deep TMHMM agreed on their predictions for 6 of 7 proteins tested from Gordonia phage Survivors, for 1 of them TMHMM 2.0 called 1 TM domain, but Deep TMHMM indicated that region was a signal sequence, so called 0 TM domains. (We won't annotate that one as a membrane protein.) I'm not sure how to evaluate the quality of the different TM predictors - at some point, it might be worth having an more guidance on that. Pam

Link to this post | posted 25 Apr, 2022 19:35

plconnerly

Thanks, Debbie! I went ahead and tried both on a small set of proteins and found an interesting difference. This preprint https://doi.org/10.1101/2022.04.08.487609 about Deep TMHMM mentions issues with signal sequences being erroneously called as transmembrane domains. Although TMHMM 2.0 and Deep TMHMM agreed on their predictions for 6 of 7 proteins tested from Gordonia phage Survivors, for 1 of them TMHMM 2.0 called 1 TM domain, but Deep TMHMM indicated that region was a signal sequence, so called 0 TM domains. (We won't annotate that one as a membrane protein.) I'm not sure how to evaluate the quality of the different TM predictors - at some point, it might be worth having an more guidance on that.
Pam

Link to this post \| posted 25 Apr, 2022 19:40
debbie	Any time one transmembrane domain is found, it is questionable. So we recommend using a second program. that usually takes care of the discrepancy. Thanks for doing the quick experiment.

Link to this post \| posted 01 Mar, 2023 19:26
edgingtonn1	DeepTMHHH keeps timing out (all day) both at the website (for my students and myself), and on the command line… Is it just me? -Nick SCSU, New Haven CT

Link to this post \| posted 01 Mar, 2023 21:55
cdshaffer	Not working for me on web page either. I did not test command line, but the error log for my web run shows that the program was not able to initialize at the biolib cloud service. Unfortunately, the web page and the command line both submit to biolib so if biolib is down not much can be done until they fix the service. I cannot seem to find any way to get the status of the biolib cloud service so if anyone knows how to that, would love to hear it. There is a way to run the whole analysis locally but it requires access to specialized GPU hardware to do the analysis so we are really just stuck waiting for biolib service to be restored. Hopefully they are aware of this issue and are working on a solution, again can't find a status page anywhere, so hard to know.

Link to this post | posted 01 Mar, 2023 21:55

cdshaffer

Not working for me on web page either. I did not test command line, but the error log for my web run shows that the program was not able to initialize at the biolib cloud service. Unfortunately, the web page and the command line both submit to biolib so if biolib is down not much can be done until they fix the service.

I cannot seem to find any way to get the status of the biolib cloud service so if anyone knows how to that, would love to hear it.

There is a way to run the whole analysis locally but it requires access to specialized GPU hardware to do the analysis so we are really just stuck waiting for biolib service to be restored.

Hopefully they are aware of this issue and are working on a solution, again can't find a status page anywhere, so hard to know.

Link to this post \| posted 02 Mar, 2023 14:50
chg60	@Chris The commandline version of biolib will submit the jobs to the DTU servers using the same API as their site uses, so in theory there's no advantage to using the biolib commandline unless you have a NVIDIA RTX 2080 GPU (or better) and a correctly setup environment. However, you can run DeepTMHMM locally even without a graphics card if you write a little Python script that imports from biolib. It takes a couple of minutes per gene, so it's not a scalable solution, nor is it a viable long-term option, but for one genome at a time should circumvent whatever (hopefully temporary) issues are being experienced on DTU's backend. I have such a script written if you're interested in trying it out, but I don't have it on this computer so won't be able to share it until later today. -Christian

Link to this post | posted 02 Mar, 2023 14:50

chg60

@Chris

The commandline version of biolib will submit the jobs to the DTU servers using the same API as their site uses, so in theory there's no advantage to using the biolib commandline unless you have a NVIDIA RTX 2080 GPU (or better) and a correctly setup environment.

However, you can run DeepTMHMM locally even without a graphics card if you write a little Python script that imports from biolib. It takes a couple of minutes per gene, so it's not a scalable solution, nor is it a viable long-term option, but for one genome at a time should circumvent whatever (hopefully temporary) issues are being experienced on DTU's backend.

I have such a script written if you're interested in trying it out, but I don't have it on this computer so won't be able to share it until later today.

-Christian

Link to this post \| posted 02 Mar, 2023 17:11
cdshaffer	OK DeepTMMHMM is working for me again but I had to create an account. It was still failing when I tried to use DeepTMMHMM as a guest. I used my github account to sign in thru OAuth but it looks like you might be able to just create an account de novo with an email address.

Link to this post \| posted 02 Mar, 2023 17:42
debbie	Thanks Chris.

Link to this post \| posted 02 Mar, 2023 18:09
edgingtonn1	Hi Christian, It hits the same biolib server on the CMD, but I was hoping that submitting them slowly overnight with a script might work, even if it was taking 5-10 minutes or longer per sequence. Instead, it just times-out, & kills the job. It looks like creating an account is the way to go – Thanks Chris! -Nick SCSU, New Haven CT

Recent Activity

Deep TMHMM?