Performance on the evaluation dataset for the methods LogitR (red), DeltaL (blue), maxF (yellow). ABC. F-score, Precision and Recall scores for different size of GO terms. DEF. The same scores against the number of annotations per protein. Smoothed splines in each subplot show fitted generalized additive models and using the R function smoothṡpline. Because a large number of points in the scatterplot coincided, we performed jittering by adding a small error term to each value e∼N(0,10−4), in order to make the maximum number of points visible.