Comparison between the performance levels of the various models.

The results from the current model are based on a 7-residue window from the protein and contain information from the sequence PSSM and global amino acid composition of the protein (for stage 2 models, the predictions from all window sizes from 1 to 7 were averaged). The p-values were computed by taking protein-wise performance scores and applying the paired Student's t-test over a set of values in the two models being compared.*These online predictions (PSIVER and SPPIDER) are based on a single model and are optimized for binding site definitions and data sets that are different from those used in this study. Although our performance appears to be higher than those of these web servers, the choice of data sets, contact definitions and performance evaluation method used were not extensively examined because the main objective of this work was to establish the point that was made in the top two rows of this table. The performance scores from the online web servers are provided only as a record (see also the results and discussion).