1 reply [Last post]
Joined: 03/04/2010
Groups: None


I'm wondering what is the protein structure prediction competition with the lowest "barrier to entry", i.e. one that does the necessary blinding that if you do well you can show you weren't "cheating", but is appropriate to hobbyists with no published results in the field?

I'm thinking of writing my own code for structure prediction, deliberately not based on any previous tools or methods because I want to challenge myself to see what I can do "from scratch". If my code works decently well on my own computer (based on predicting folds from the PDB whose structure is in no way input into my algorithm), I would release it open source. However, obviously others won't know to trust any benchmarking I do on my own (nor should they haha!), since if I have access to the PDB entries I could have given "hints" to the code, plus I wouldn't know how well it compares to more traditional approaches.

So I'd like to do some type of blinded test, and based on that decide whether it's worth even thinking about publishing. If I do really, really well I might decide to "quit my day job" (as a research assistant in a totally different field) and apply to jobs in the field of protein structure. I'd love to be the "Ramanujan (https://en.wikipedia.org/wiki/Srinivasa_Ramanujan) of protein folding", it's one of my life dreams.

As I understand really anyone can join CASP, but as I understand you need to pre-submit a methods section, and while I'd be OK summarizing the general ideas of my code publicly, since there wouldn't be any prior publications to reference so people could actually understand the method based on an outline like that, I'm wondering if it could be rejected based on that, even if the assessment of the structure itself is automated and blind. It's possible this isn't a concern, though, and I consider it likely that there are other more informal competitions out there, like "pre-qualifying rounds" for CASP if you will. Basically I want something undemanding enough of entries that somebody's dog could submit a prediction, yet where I am not assessing myself in a biased manner. Almost certainly the assessment would be mostly if not entirely automated, as I'm sure there aren't many live humans who know anything about proteins who would enjoy manually looking at umpteen very unserious predictions that would no doubt be submitted to such a competition.

Do any of you know what the world of prediction competitions looks like?

beta_helix's picture
User offline. Last seen 10 hours 36 min ago. Offline
Joined: 05/09/2008
Groups: None
CASP won't start up again until 2022

If you can automate your methods, you might try CAMEO:

The advantage of CAMEO is that it runs every week using new structures released by the PDB, although it does require you to use a server.


Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, Boehringer Ingelheim, RosettaCommons