Figure 2
From: AVID: An integrative framework for discovering functional relationships among proteins
Evaluation of stages 2 and 4 using known proteins. (A) To evaluate the performance of a simple product model in stage 2, PjAVID2 was calculated for each pair of proteins in an annotated reference set. The plot shows the fraction of protein pairs at different PjAVID2 cutoffs that share an AVID GO term. Pairs with PjAVID2 < 12.8 were not considered in stages 3 or 4. (B) A varying percentage of reference proteins was omitted from the entire training procedure and used as a test set. Functions for these proteins were predicted in stage 4, and the plot shows the success rate for correctly predicting at least one existing GO term at the highest level of annotation. The arrows indicate the expected performance for predicting new functions based on the current status of annotation of the yeast genome. Error bars show the standard deviation from 100 random cross-validation trials. MF (diamonds), BP (squares) and CC (triangles).