Ligand Similarity Complements Sequence, Physical Interaction, and Co-Expression for Gene Function Prediction
(A) The prediction of GO annotation terms grouped by evidence code and sub-ontology by individual and combined networks. The ChEBI subset consists of terms associated with the Chemical Entities of Biological Interest (ChEBI) ontology (S3 Table, S3 Fig). Error bars represent the standard error of the mean. Combining the networks improves performance substantially (average ~0.80). (B) The CoExp network performance gains with increasing sample size but with diminishing returns, especially when compared with the gains obtained by combining the orthogonal chemoinformatics network. Extrapolating the aggregation curve (orange line), we predict that we would need millions of more samples to achieve similar performance with CoExp alone as with the combined chemoinformatics and CoExp networks (orange arrow).