Fig. 3: The distributions of the performances of the protein function prediction approaches tested in this work.

This performance was measured in terms of the F_max score for each of the 2,139 GO terms. For EI and the individual modalities, the score for the best performing heterogeneous ensemble method for each GO term is shown here. For the Mashup and deepNF early integration methods, the score of the best local modeling algorithm for each term is shown.