Genome-Wide Prediction and Validation of Peptides That Bind Human Prosurvival Bcl-2 Proteins
Figure 5
Predicting peptide binding to the 5 Bcl-2 receptors.
The first benchmark (A) included 366 interactions (KD<1 µM) and non-interactions (KD>10 µM). Four models were evaluated with respect to their ability to correctly classify each example, as a function of the score cutoff used for prediction. The second benchmark (B) included 180 comparisons of one receptor binding a peptide (KD<1 µM) and another receptor not binding that same peptide (KD>10 µM). The difference in scores for a peptide binding to two receptors was used to predict the binding preference, and agreement with experiment was evaluated as a function of the score difference cutoff. The “PSSMSPOT+STATIUMSC” score is the average of the Z-scores of the two models for a given receptor. Values in parentheses report the area under the curve (AUC) for each method. For details, see the Methods section.