I need an explanation about Rosetta de novo structure prediction and fragments picking

Question

[Palming]

Hello everyone,

I am currently trying to work with Rosetta. I have a peptide sequence and I would like to predict its structure using de novo structure prediction. I have Rosetta installed and I need to do this with Rosetta (so please, do not suggest any webservice such as Robetta). I tried following the Rosetta documentation: the de novo structure prediction tutorial states that required input files are the sequence in the fasta format and 9mer & 3mer fragments files. To obtain the fragment files, the turorial states:

There are multiple ways to obtain those fragment files. You can generate them on your local machine. For that to work you need additional software installed. How to do that is explained in a separate tutorial for fragment picking.

However, the fragment picking tutorial says you need three input files to generate fragments: the fasta sequence (which I have) ; the psipred .Ss2 file (which I have) and the structure (which, of course, I don't have). I have tried running the script for fragments picking: it works well when an input structure is given, but it doesn't work if there is no structure.

Therefore, here is my question: how can I do an ab initio structure prediction if I need fragment files whose generation requires a structure ? How do I generate fragments if I don't have a structure ? Or how do I do ab initio structure prediction if I don't have fragments ?

I am fairly new to Rosetta and the concept of fragments, so if I am missing something basic please tell me.

[shahram.mesdaghi]

Hi I have the same problems as you had. So how did you 'generate a fragment library from all available structures'? Thank you!

Answer 1

[Palming]

If anyone is ever interested, I finally got the answer to my question :

Contrary to what h.mon suggested, you are not supposed to make a homology (or threading) model of your protein, then generate fragments from this model, and then use these fragments to make an ab initio model in Rosetta.

You are supposed to generate a fragment library from all available structures in databanks (e.g. from the PDB) and then generate your ab initio model. Which makes more sense.

Feel free to correct me if I'm wrong, or to add any relevant information for future reference.

[dokto20m]

How did you generate a fragment library from all available structures in databanks?

Answer 2

[h.mon]

You can obtain a pdb model if you check the BioSerf option at the PSIPRED server.

BioSerf Output

If you provide a valid MODELLER key you will have been able to run a BioSerf job. BioSerf is a fully automated homology modelling service which integrates PSI-BLAST, HHBlits, PSIPRED, GenTHREADER and MODELLER. The final output is a PDB file which can be viewed by clicking the BioSerf tab on the results page.

[Palming]

Thank you for your answer. What doesn't make sense to me is generating fragments from a model. If I make fragments from a homology model, then the ab initio model will highly depend on the quality of the original model, won't it ? In that case, why even bother making an ab initio model ? Can it be better than the homology model the fragments are based on ?