Fig. 4
From: Machine learning for discovering missing or wrong protein function annotations
Procedure used to update each FunCat dataset. The sequence IDs are extracted from the 2007 dataset, and used to query new annotations using UniProt. A hierarchy (subset of FunCat) is built using the new annotations. Finally, the old annotations are removed, and the new dataset is created by concatenating the new annotations with the feature vector and IDs