Submitted: "Hong Hung" <lhhung@u.washington.edu>
Date:  Mon, 15 Jul 2002 12:32:20 -0700
 
 

    PsiCSI uses neural networks to translate chemical shifts to secondary structure information and combine it with sequence based prediction algorithms (PsiPred). For a rigorously jack-knifed set of 92 proteins, PsiCSI made 36% fewer errors than CSI, achieving a sustained 3-state accuracy of 89%.  In addition, because of the sequence based component, the method remains effective with sparse and incomplete chemical shift data.

    A webserver is available at  http://protinfo.compbio.washington.edu/
 

---