Computational Protein Function Prediction: Framework and Challenges

Large scale genome sequencing technologies are increasing the abundance of experimental data which requires functional characterization. There is a continually widening gap between the mounting numbers of available genomes and completeness of their annotations, which makes it impractical to manually curate the genomes for function information. To handle this growing challenge we need computational techniques that can accurately predict functions for these newly sequenced genomes. In this chapter we focus on the framework required for computational function annotation and the challenges involved. Controlled vocabularies of functional terms, e.g. Gene Ontology, MIPS functional catalogues, Enzyme commission numbers, form the basis of prediction methods by capturing the available biological knowledge in the form, suitable for computational processing. We review functional vocabularies in detail along with the methods developed for quantitatively gauging the functional similarity between the vocabulary terms. We also discuss challenges in this area, first pertaining to the erroneous annotations floating in the sequence database and second regarding the limitations of the functional term vocabulary used for protein annotations. Lastly, we introduce community efforts to objectively assess the accuracy of function prediction.