For example, it has often been noted that U-turns typically contain very low-scoring amino acids when Show Relative Score is turned on. A U-turn is frequently the lowest-scoring area of a puzzle, but this does not necessarily mean that it is always the most suitable area for improvement. This is because while the local area may score poorly, the presence and location of the U-turn may allow other, quite large areas to score very well -- and adjusting the U-turn may throw these other areas out of alignment. A U-turn between two well-positioned sheets sharing many hydrogen bonds, with hydrophobics hidden deep within the protein, will cause these sheets to score very well. Adjusting the U-turn with the goal of raising its score may or may not yield any additional points, depending on how these adjustments affect the global structure.
With that said, U-turns are still very often good targets for rebuilds, local wiggles, and the Manual Rebuild strategy. Locking the sheets or helices that they connect, and using local-only adjustments until the protein stabilizes, are frequently advisable if you want to preserve the good characteristics of the connected pieces while exploring alternate U-turn configurations.
Some articles regarding U-turns:
http://bioinformatics.oxfordjournals.org/content/20/16/2751.long is a free article (PMID: 15145798 [PubMed - indexed for MEDLINE]) about using neural nets for predicting beta turns. 2 webservers for this are at:
- http://bioinformatics.uams.edu/mirror/betaturns/ [mirror site]