You are here

Data Check for a newb

3 posts / 0 new
Last post
Data Check for a newb
#1

Dear Rosetta Community,
 
I am designing a protein and this is the result of my abinitio fold. I generated 24,000 structures and the SCORE vs RMSD plot is attached.
 
As you can see, I do not have a "clear" funnel shaped plot, but I do have several low scoring and low RMSD structures.
 
* Is this considered a funnel shaped graph? given that I have "some" and "not a lot" of low scoring low RMSD structures?
 
*is this protein worth expressing? or will it just prefer to fold into a low scoring high RMSD structure?
 
*should I try to modify it to get more low score low RMSD hits? or this is enough?
 
I hope I am communicating my idea and questions clearly, I am still new to rosetta and working to gain experience.
 
Thank you.
 
Ac Research
AttachmentSize
Score vs RMSD abinitio plot245.48 KB
Post Situation: 
Sun, 2017-03-12 23:42
ac.research

Koga et al. (https://doi.org/10.1038/nature11600) as well as many of the other de novo protein design papers have examples of what the energy landscape (the score-vs-rmsd plot) for sucessful designs look like. You can compare your results with theirs.

One thing you might want to try is to diagnose is if you're doing sufficient sampling in the abinitio stage. Generally what's done with forward folding is to also run a small relax run with the designed protein, to give a sense of where the folded form is on the score-vs-rmsd plot. (These are the green blobs in the Koga et al score-vs-rmsd plots.) If the relax is still much lower in energy than the forward forlding runs, it might just be that you haven't sampled enough in ab inito to find the low energy fold. On the other hand, if the 12 Ang strucutes are the same energy as the relaxed structures, you know your design doesn't have a good funnel. (Even if the relaxed run is lower in energy, you might want to focus on a different design, one which has an easier time finding the low energy structure.)

From my quick glance, Your current plot doesn't look so good. I might recommend trying a different design. Keep in mind that when desiging proteins, the general principle is to design many, many different proteins compuationally, and then throw out most of them. I can tell you that Koga et al. put a *large* number of structures through forward-folding before selecting just a few to put into the wet lab. Much of a design is a numbers game.

That's not to say you couldn't try this protien in the wetlab, but keep in mind that the point of forward folding is to identify designs that clearly like their folded state. If you don't have as clear a funnel as Koga et al, then you're likely to have a much lower success rate. (The concept behind forward folding is that computational time is cheap compared to wet lab time, so it pays to spend computational resources making and discarding a large number of designs which may be "iffy" in order to focus wet lab efforts on the fewer structures with a higher probability of being good.)

As mentioned, you may want to go back to the design stage, and pick out different designs. One thing which might help your throughput is to take a look at fragment quality. Before running the forward folding, look at the individual fragments which are predicted from just the sequence. If they don't match the designed structure well (either in aggregate or just looking at the closest fragment), then your abinitio run is likely not to give you decent results. You can focus your forward folding runs on just those designs with good fragment scores. (Or more precisely, those sequences which are predicted to give you secondary structures which you desire.)

 

 

 

Mon, 2017-03-13 08:57
rmoretti

rmoretti,

 

Would it be a bother if i ask to communicate with you by email? so i can send you some of my results? I read the koga paper several times and started to get better results, but i might still have an issue or two.

Fri, 2017-03-31 07:36
ac.research