Differences in the predictions by I-TASSER web server and standalone package

This forum is shown on the index page along with all topics.

Moderator: robpearc

sambitmishra
Posts: 3
Joined: Mon Apr 18, 2022 7:12 pm

Differences in the predictions by I-TASSER web server and standalone package

Post by sambitmishra »

Hi,

I used I-TASSER to make predictions on a protein of interest and observe significant differences in predictions made by the web-server and the standalone package.

The predictions made by the webserver are shown below. As you will notice, the best predicted structure (Model 1) has a TM-score of 0.56 and C-score of -1.25, both of which are good.
Webserver prediction
Webserver prediction
webserver_prediction.png (238.82 KiB) Viewed 502 times

The following image shows predictions for the same sequence from the standalone package installed on the Biowulf cluster.
Standalone prediction
Standalone prediction
standalone_prediction.png (156.44 KiB) Viewed 502 times
As you will notice, the best model from Biowulf has a TM-score of 0.35 and C-score of -3.24, which is significantly different from the values observed for the webserver models. Even you will notice that the PDB templates are different for the two cases. I am not sure why. It does look like the webserver version of I-TASSER gives superior results than the standalone version.

Please advise.

Thanks,
Sambit Mishra.
zcx@umich.edu
Posts: 24
Joined: Wed Nov 03, 2021 5:59 pm

Re: Differences in the predictions by I-TASSER web server and standalone package

Post by zcx@umich.edu »

The online web server contains some component that we unfortunately cannot redistribute with the standalone package due to license issue. For example, the template identification by LOMETS uses HHsearch, which is only used by web server but not standalone package. In most cases, this does not have a big impact; but in your specific case, it does lead to some differences in templates and final model confidence.
afif
Posts: 1
Joined: Thu Apr 21, 2022 7:40 am

Re: Differences in the predictions by I-TASSER web server and standalone package

Post by afif »

Hello,
I'm from the NIH HPC team and Sambit also raised this issue with us. So is there a way around the issue? If we separately obtain HHsearch, can we set it up so that LOMETS uses it? And is there anything else we could set up separately?

thanks and regards
Afif
sambitmishra
Posts: 3
Joined: Mon Apr 18, 2022 7:12 pm

Re: Differences in the predictions by I-TASSER web server and standalone package

Post by sambitmishra »

Hi there,

Also wanted to ask: Does the I-TASSER webserver set the command line option -light to True when running calculations?

Thanks,
Sambit.
xiaogenz
Posts: 139
Joined: Sun Apr 25, 2021 12:02 am

Re: Differences in the predictions by I-TASSER web server and standalone package

Post by xiaogenz »

afif wrote: Thu Apr 21, 2022 7:44 am Hello,
I'm from the NIH HPC team and Sambit also raised this issue with us. So is there a way around the issue? If we separately obtain HHsearch, can we set it up so that LOMETS uses it? And is there anything else we could set up separately?

thanks and regards
Afif
Hi,

Yes! You can search the templates by HHsearch and provide them to the I-TASSER server using Option I.
xiaogenz
Posts: 139
Joined: Sun Apr 25, 2021 12:02 am

Re: Differences in the predictions by I-TASSER web server and standalone package

Post by xiaogenz »

sambitmishra wrote: Mon Apr 25, 2022 4:55 pm Hi there,

Also wanted to ask: Does the I-TASSER webserver set the command line option -light to True when running calculations?

Thanks,
Sambit.
Thank you! The I-TASSER server uses the standard version rather than the light version.
Post Reply