Results of our SNOB algorithm vs. the algorithm published in [14].

The “False Negatives” column shows the total number of false negatives produced by each algorithm across all 101 BP categories examined in the paper, as determined by BP data collected by the authors two years after the training data. The “Negative Frequency” column shows the average number of times any gene was selected as a negative example for different function categories, if it was selected at all (a higher number means the same proteins are selected as negative examples across more categories). The “Avg Enrichment P-Value” column is the metric the authors used to evaluate their function predictions, with a lower value indicating better performance (see [14] for details).