That is very, very impressive, whether I understand it or not. It is a pleasure to work on such a project.

@ErikaT: Thanks for the information! Interesting read!

@sptrog1:
From where do you know that they do not run working units on their own machines? Are there any public information available?

Well, 275k unique proteins is much compared to the knowledge we had in this space before the project. You can read the protein you calculate in the work unit. A work unit description is like MIP1_00269443_1000_0. The first number represents the protein and the second number the experiment for the protein. Usually around 2000 experiments/calculations are needed to predict a protein. This conforms with the given data: 436 millions results / 275 thousand proteins = 1658

It seems strange to me, cause they are using Rosetta code
Or not?

Let's remember that our computing is determining protein structure, not predicting protein function, which is what the new technique does.

What I am unclear on is how this statement can be true given the extent of work done at the Rosetta@Home project:
"State-of-the-art methods to date relied only on protein sequence information...because researchers never had access to enough protein structure data to make such predictions widely applicable....We tackled...the data availability problem by enrolling your help through the Microbiome Immunity Project and generating a collection of 3D protein structural models of an unprecedented scale."