Current protein results

[MAD-ness]

I have been watching the RMSD of this new protein, which is the largest we have run so far. As of now, there are 8 returned structures that have RMSD of less than 6A, if I am reading the stats page correctly. We are about half way through the run of this current protein, so it seems likely that by the end of the run there will more than 10 returned structures with an RMSD of less than 6A.

Here is a little quote I ganked from the science section of the distributedfolding.org site:


However, an RMSD of 6A or less for a protein of fifty or more residues indicates that the general topology is correct. At this level of resolution, crude scoring functions can be used successfully to help identify the native fold.

Here is another (from the following paragraph):


We have estimated that for a typical medium-sized protein of about 120 residues, by sampling on the order of 1 - 10 billion probabilistic conformers a structure within this 6A tolerance could be sampled, and would probably have the correct topology (i.e. fold) and one of the lowest energies out of the structures generated. If this hypothesis is correct, this means that the structure of almost any 120 residue or shorter protein could be predicted ab initio, i.e. without any prior knowledge about the protein.

I am sorry if these quotes are somehow misleading or if my understanding of them is mistaken, I am not attempting to confuse matters.

It looks as if the results for smaller proteins are going to be under 6A (I think the last two protein's also had RMSD well under 6A), which would seems to indicate that the algorithm and the project are both of some immediate value. Not being a biologist, I am not sure how many proteins are in the 50 residue range and if any of these 'smaller' proteins are involved in any of the reactions that interest the medical field, but it seems to be a case of "so far, so good."

Are the results on the 3 proteins which have been computed (including this half finished one) in line with your predictions and do you expect the early results to yield information which will enable you to refine and improve the algorithm for future runs?

Most of us are not able to follow the high level (or low level for that matter) science involved here, but the information I can glean from the science page as well as the TraDES page is very interesting to read.

Thanks.

[jkeating]

Good comments MAD-ness. I'd like to hear the answer to that also!

[Brian the Fist]

Well, we will not be able to draw any true scientific conclusions until we have the time to properly analyze the data we have collected so far (hopefully within a several weeks). However, as far as the RMSD goes, the best structures generated or roughly on par with what we expected. The bigger structures will be more interesting as they are generally much "harder" to get within our somewhat arbitrary cutoff of 6A. If you look at the current 5.6A best structure in Cn3D and the Native structure (you can do this from the main stats page after installing Cn3D) you can see that indeed the topologies are approximately the same though the precise fold differs slightly. It remains to be seen (when we analyze the data) how well the energy scores can pick out these low RMSD structures (since in a novel protein of unknown structure, we do not know beforehand the RMSD - that is, we must go by the energy alone to pick out the best structures). The 6A structures are not precise enough to start designing drugs to target these molecules, but they are close enough to either combine with experimental data to get the true structure more quickly, and can possibly be refined through other computational techniques to bring them closer to the "correct" structure. We will be investigating different possiblities as the project proceeds.

Thanks for your interest and I hope I answered some of your questions at least.

[MAD-ness]

Yeah, that was great.

I was intending to ask more about general information rather than journal published or doctoral thesis quality technical information. ;)