Does anyone have experience generating pdb structures with Phyre and ITasser online tools. The results generated from each given the same amino acid sequence input are very different and I am wondering whether or not this is a usual experience. I know ITasser was the number 1 rated in the CASP trails, but should the results really be this disparate?
$\begingroup$
$\endgroup$
1
-
$\begingroup$ You should always pass your models through a model quality assessment program to evaluate reliability, and select the best one. $\endgroup$– DavidmhJun 1, 2017 at 14:18
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
1
I'm less familiar with Phyre, but I-TASSER is a really sophisticated system that takes the results of a search using multiple threaders and plugs them into an ab initio simulation which tries to minimize the energy of the models by sampling many possible 3D conformations, which I don't think Phyre does.
https://en.wikipedia.org/wiki/I-TASSER#/media/File:I-TASSER-pipeline.jpg
Compare with a similar workflow schematic for Phyre:
http://www.nature.com/nprot/journal/v10/n6/images/nprot.2015.053-F1.jpg http://www.nature.com/nprot/journal/v10/n6/fig_tab/nprot.2015.053_F2.html
Structure prediction still has a long way to go, and you'll always get better results if there are close homologues available in the PDB, but given the consistent high performance of I-TASSER in CASP I would treat those results as more significant. That said, it can't hurt to consider multiple answers.
Edit: included blmoore's link to second half of Phyre protocol schematic
answered May 20, 2017 at 4:11
-
1$\begingroup$ You're only linking to the first half of the phyre2 workflow: nature.com/nprot/journal/v10/n6/fig_tab/nprot.2015.053_F2.html $\endgroup$– blmooreMay 24, 2017 at 13:05
$\begingroup$
$\endgroup$
This may not be the whole answer to the question, but one key difference between Phyre and ITasser (I think I'm correct in saying), is that Phyre only returns regions of the model it considers to be sufficiently well modelled, so it may not be a full length model. ITasser on the other hand, always generates full size models.
As one of the other answers alludes to, it is very dependent on your protein of interest. If it isn't well represented in the databases, the initial threading stages will not yield good models, and the subsequent steps all suffer as a result.