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?
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.