CASP8 Results: Human Vs. Servers
The 8th community wide experiment on the critical assessment of techniques for protein structure prediction ,or CASP8 for short, has ended a couple of months back, and the results are in. In this CASP, 112 human expert groups were registered and 121 automatic prediction servers. 128 targets were released for prediction, generating a total of 80,560(!) submitted models.
According to the CASP website, for the human expert groups on 71 template based modeling (TBM) and free modeling (FM) targets, the top three groups were:
For the server’s automatic predictions (164 TBM & FM) the top rankings were:
Many other assessments exist (Zhang,Baker,Grishin,McGuffin,Cheng) which show quite consistently according to different measures that the Zhang server is ranked first amongst servers, while the Baker group is ranked first for the human/server targets. Is there still some human intuition in protein modeling that can not be formulated into a server ?
Another anecdotal indication for this trend are the FoldIt! players results in CASP8 : for 7 targets, players or groups were ranked amongst the best 3 predictions, and for one target they actually predicted the best model (out of 77 entries).
So the next time you need to model a protein, will you use a server ? or operate a modeling software yourself ?
