Tuning, belatedly.

Well, it looks like we didn’t do overly well at Casp – well still in the middle of the pack for most predictions.  Not surprising, as we get better so does the competition.  However – despite the lowish ranking there isn’t a group that we didn’t beat at least once – so we don’t produce rubbish models.   IMHO there isn’t a big difference between decent servers for structures with 40% or higher sequence identity.

There is an interesting idea for modeling potentials that comes from the Skolnick group.  Basically take the hydrogen bond function from dssp and use it with Van derWaals terms to describe the long range interactions.  Neat and simple – so of course I’ve implemented it.

My preliminary results are sort of interesting.  First, it is not a folding potential – even though it looks formally like converting a continuum problem into an Ising problem.  When I run molecular dynamics with it, it can easily find very non-native folds that are much lower in energy than the native fold.  However, it is a good refinement potential – it tightens up the  structure and results in better secondary structure and slightly lower RMS deviations as well as better fractions of the structure superimposed  at 1A and 4A cutoffs.   It probably would have moved the performance of my server up a notch.

Written by Rob in: engineering,laboratory practice |

No Comments »

RSS feed for comments on this post. TrackBack URL

Leave a comment

Powered by WordPress | Aeros Theme | WordPress Themes