PMC

Rate of discovery of new eukaryotic PPI data has slowed.

Web interface and output of Struct2Net. (A and B) Web server entry page. (C) A query option for either a quick-but-approximate approach (using orthology over pre-computed predictions from yeast, fly and human) or a full-blown computation using the Struct2Net algorithm. (D) Example of an output page when choosing to thread pairs of sequences onto all templates. Confidence scores for a potential interaction are displayed along with associated template–sequence alignments and threading details.

Sensitivity versus specificity. The prediction algorithm can achieve 60% sensitivity while maintaining 75% specificity as measured on the test set. Here, sensitivity = (true positives)/(true positives + false negatives) and specificity = (true negatives)/(true negatives + false positives). We constructed a training set and test set of positive and negative examples from yeast and fly, using criteria we have developed to identify high-confidence positive and negative examples of PPIs (see the website FAQ for details). After training the logistic regression model on the training set, its performance was measured on the test set.