Fold recognition using sequence and secondary structure information

We applied a succession of sequence search and structure prediction methods to the targets in the fold recognition part of the CASP3 experiment. For e ach target, we expanded an initial sequence space, obtained through PSI-BLA ST, by searching for statistically significant relationships to low-scoring sequences and then by searching for conserved sequence patterns, We then d ivided the proteins in the sequence space into families and built an alignm ent hierarchically, using the multiple alignment program MACAW. If no signi ficant similarity to a protein of known structure was apparent at this poin t, we submitted the alignment to the Jpred server for consensus secondary s tructure prediction and searched the structure space using the secondary st ructure mapping program MAP, Failing this, we compared the structural prope rties that are believed we recognized in the aligned proteins to the folds in the SCOP database, using visual inspection. If all these methods failed to uncover a plausible match, we predicted that the target would adopt a no vel fold. This procedure yielded correct answers for seven of twenty-one ta rgets and a partly correct answer for one, A retrospective analysis shows t hat automating the sequence search procedures would have represented a sign ificant improvement, with at least three additional correct predictions. (C ) 1999 Wiley-Liss, Inc.