Hi all,
I’m trying to do a homology model using multiple templates taking as reference the RosettaCM tutorial (https://www.rosettacommons.org/demos/latest/tutorials/rosetta_cm/rosetta_cm_tutorial).
I have some doubts regarding the use of some weights for each of the stages.
First, taking a look in the tutorial and since the protein I’m using is not from the membrane I’m wondering if I should use the stage1.wts, stage2.wts and stage3.wts files from the tutorial or the ones that I’ve seen in other references (maybe older):
<stage1 weights=score3 symmetric=0>
<Reweight scoretype=atom_pair_constraint weight=0.5/>
</stage1>
<stage2 weights=score4_smooth_cart symmetric=0>
<Reweight scoretype=atom_pair_constraint weight=0.5/>
</stage2>
<fullatom weights=talaris2014_cart symmetric=0>
<Reweight scoretype=atom_pair_constraint weight=0.5/>
I know the difference between those files are in the given values for the energy terms so my question is if is there any preference (i.e if you have a template with high % of similarity with your protein of interest) or the performance is not so different. Furthermore, I wonder if it would be necessary to add the fragments files (3 and 9 residues length) if there are some region without template (maybe a disorder domain) although the similarity is high in other parts. I don’t know if this can add some noise to the results since it is more used with ab initio.
Also I want to ask you if you think it would be necessary to add a relax step as I‘ve seen on the example of the hybridize mover (step 2) (https://www.rosettacommons.org/docs/latest/application_documentation/structure_prediction/RosettaCM).
But the relax step is missing on the xml file obtained from the setup_Rosetta.py while it seems this step is included on the rosetta_cm.options when creating the files with the python script:
-relax:min_type lbfgs_armijo_nonmonotone
-relax:dualspace
-relax:jump_move true
I am not sure if I should just add to the hybridize step on the xml script or use the flag with the roseta_cm.options.
Thank you so much for your help and advice,
Fragments: you should always use fragments. It's important as a source for model diversity within the well-templated regions, and the only way to fill in the poorly templated regions.
Relax: I think this step is optional. If you have CPU time to burn, go for it.
Scorefunction: I don't know. I think you'll see relatively small differences.
Which references are you seeing those other weights in? (And what do they say about the source of those weights?) The tutorials/demos don't always keep up with the latest in developments, so it may indeed be that those are better. -- Spot checking things, I notice that the score3.wts fullatom scorefunction in the tutorial is basically the old score12 scorefunction, whereas the ones you post are talaris2014 based. If you're not doing membrane proteins, I'd go with talaris2014 over the score12-based ones.
You probably want to do a relax step after your homology modeling stage. From the people who do RosettaCM around here, it's very hard to tell good models from poor models directly out of the low resolution stages - you do need a relax step to clean things up. Now whether that relax step happens in the Hybridize mover itself, or happens externally, later, that's up to you. The tutorial is written as having a separate relax step because there was (still are?) issues with imposing disulfides during the relax phase. Running it separately later means you can pass a -in:fix_disulf to impose the known disulfide bond on the protein during the high-resolution relax stage. -- the flags you posted are reasonable. The -relax:dualspace option is something you definitiely want to include, as it will reoptimize bond lengths and angles during relax, and the low resoultion stages of Hybridize tend to often produce odd geometries there. (One of the reasons why it's hard to assess the proteins prior to a relax step.)
Fragments: Fragments are definitely a good idea, especially if you have untemplated regions. (But ideally you should include them even with complete template coverage - there isn't necessarily a guarantee that the template structure is adequate to model the homolog's structure.)
Relax: I agree with Rocco there; I was assuming you had relax running inside hybridize, which is how it works in the workflow I have access to. Take my comment as based on possibly faulty assumptions.
I would first like to thank you for the detailed responses.
Answering R.Moretti, I extracted the weights from the setup_RosettaCM.py script inside the /rosetta/tools/protein_tools/scripts. (altough I modified the talaris2013_cart for the talaris2014_cart, more recent version).
Then, I hope this is not a stupid question but for the relax protocol is it enough to put the flag (-relax:dualspace or other relax options) or also it is necessary to put in the xml protocol,(code below) . Or even both since for me it is not clear that just writing on the xml file it will know that I want the dualspace relaxation (internal coordinates + cartesian optimization)
<MOVERS>
<FastRelax name="relax" scorefxn="fullatom" />
</MOVERS>
<PROTOCOLS>
<Add mover="hybridize" />
<Add mover="relax" />
</PROTOCOLS>
<OUTPUT scorefxn="talaris2014" />
</ROSETTASCRIPTS>
Again, thank you so much for you time and a really nice program
Either putting `-dualspace` on the commandline or putting `dualspace="true"` in the XML should work. The value in the XML will override the value on the commandline, but if you don't specify it in the tag, it will take it from the commandline.