[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [modeller_usage] is LOOP threading a common result?



Starr Hazard wrote:
I am using Mod8v1 on a MAC G5 running OSX 10.3. In a protein with about 500 aa there was about a 100 amino acid region which did not match the single available template. I asked Modeller to make 1000 loops in the gap region. The happened and after a few fits I got the DOPE analysis of all 1000 loops complete. I picked the best 20 loops (lowest overall DOPE scores;minimal plots in the loop region) and examined these structures. In every one of the best twenty AND in the worst loop model, I see regions where the routines have threaded the loop though the nonloop regions of the original model. In words, suppose I have two antiparallel beta sheet strands. The backbone of the loop sometimes passes between these sheet stands or through the turn loop at the ends of the sheets before coursing away from the existing protein.

Do the loop generator routines pay any attention at all to the
nonloop regions?

With loop modeling, you are essentially doing protein folding, so you will be extremely lucky to accurately model a 100 residue loop. You should try to find a template, even at low identity, for at least part of the region.

When loop modeling, Modeller will first optimize the loop without considering the nonloop regions, and then repeat the optimization, enforcing soft-sphere interactions with all nonloop atoms within 7 angstroms. It is entirely possible that your loop is being threaded into the nonloop region during the first phase, and the second optimization cannot recover it. If you really want to model this region as a loop (see disclaimer above) then you can adjust this behavior by overriding the build_seq() routine of the loopmodel class. Sampling is also likely to be an issue in this case, so you will need to use more optimization and build more models.

	Ben Webb, Modeller Caretaker
--
             http://www.salilab.org/modeller/
Modeller mail list: http://salilab.org/mailman/listinfo/modeller_usage