I found that the "psipred" installed at three different places generated different results for one sequence (Rad59, please focus on the 40 aa at its N-terminus). Is that because their version? Which "psipred" is more reliable?
1. "Fragment Libraries" server (v2.6): http://robetta.bakerlab.org/downloads/fragments/22521/
2. the SYMPRED server (v ?): http://zeus.few.vu.nl/jobs/aeb9f401c184595c85e6ef7aaef6defe/
3. A local installed Psipred copy (v. 3.2 ?).
The secondary structure prediction will affect the fragment libraries. How much do the fragment libraries affect the final AbinitioRelax prediction?
'Which "psipred" is more reliable? '
This is really a question for the developers of psipred more than us. I would hazard a guess that the most recent version is "best".
I think the secondary structure libraries matter quite a bit to the final result. I'm not aware of anyone benchmarking slightly-different fragment inputs for folding a known structure to see which one comes out best.