Hi, I am studying a number of proteins, for which I am obtaining .pdb models via the ROBETTA online modeling service. I use this ab-initio method since no suitable templates are available for homology modeling.
My answer is: from the five models I am given for each protein, ¿how can I know which one should be closest to the real thing? I know of a couple of sites as QMEAN and ERRAT, which are used for model checking. I could just use the QMEAN results, but it doesn't feel enough to decide that a model actualy is getting close to the real thing.
Which would be the most acceptable approach to get closer to the real protein structure without getting into NMR, crystallography or other bench experiments? If I can get a plausible candidate, I would like to try and infer what and/or how the protein works. An idea that comes to my mind is: Could I for instance look for homologues from other species, and see if one candidate structure is consistently obtained among the obtained models?
I'd like to be more specific, but at the point I am now I guess I just can ask: any suggestions? Where can I get the working knowledge to effectively address this problem?
Thanks a lot in advance,
Karel
+1 for a really cool non-NGS question. Wish I could help but I don't know much about proteins.
Have you looked at the Robetta FAQ? Also, I think that question may be more well suited for the Rosetta mailing list.
As for getting a more plausible candidate... What kind of protein are you working with? How much do you know about the protein? And how have you reasoned that no suitable templates are available?
Actually most of what I know comes from sequence alignments of homologues, and from launching the sequences against bioinformatic property prediction tools. There is also the chance that it is a membrane-interacting protein (to make things worse). As for the absence of good templates, I can tell from the results obtained with homology modelling, which seemed to yield non-sensical results.
The best results I have obtained for ab-initio approaches (evaluating them with QMEAN) were with the mentioned Robetta service, as well as with I-Tasser, QUARK and CABS-fold. Curious thing, is there seem to be considerable divergence among models obtained with those different services and Robetta. On the other hand, the latter three tend to give results more similar among themselves, but generally yielding somewhat poorer QMEAN scores for my proteins.
Cheers,
Karel
The question still comes down to: How did you pick your template? I'm not quite sure I understand your answer. All of your prior knowledge of the protein in question comes from sequence alignments of homologues? That's not a lot of prior information.
The reason why I keep asking for a template is that, as someone else mentioned, without a template you're very unlikely to get good results. From what I've heard, precise homology modelling even with a very good template is extremely difficult.