T0542 - servers beats experts

[arneelof]

I found the results for T0542 particularly interesting.

The 10 best models are from servers and the best manual prediction has a GDT_TS of 40 compared to 60 for the best server models

Would be fun to know how these models where made

Arne

[djones]

Hmm - only seems to be true of the complete chain, not the separate domains.

Presumably the best servers just happened to pick the right full length template for the whole chain i.e. one with the two domains oriented correctly for
the target.

Also, it's a server only target, so many humans either didn't try it at all or didn't put much effort in.

[arneelof]

Good point

[jianlin.cheng]

Indeed, our server MULTICOM-CLUSTER used quite a few templates including 3ILVA 2E18A 1XNGA 3FIUA 1KQPA 2PZAA 3DPIA 1WXIA, 1IH8B, 2PZBA, 1ERZA. 3LVA covered the entire target and 1ERZA covered the first domain. All other templates covered the second domain. MULTICOM-CLUSTER did pretty well on the second domain and OK on the first domain. 3LVA might be the best template. Other templates may also contribute to the good quality of the second domain.

[jianlin.cheng]

Continued with my last post. The alignment file used by MULTICOM-CLUSTER is http://sysbio.rnet.missouri.edu/casp9_h ... 0542_1.pir, which was used to generate the first model: http://sysbio.rnet.missouri.edu/casp9_h ... 0542_1.pdb.

--Jack