Automatic Generation of Functional Annotation Rules Using Inferred GO-Domain Associations

International audience; The GO ontology is widely used for functional annotation of genes and proteins. It describes biological processes (BP), molecular function (MF), and cellular components (CC) in three distinct hierarchical controlled vocabularies. At the molecular level, functions are often performed by highly conserved parts of proteins, identified by sequence or structure alignments and classified into domains or families (SCOP, CATH, PFAM, TIGRFAMs, etc.). The InterPro database provides a valuable integrated classification of protein sequences and domains which is linked to nearly all existing other classifications. Interestingly, several InterPro families have been manually annotated with GO terms using expert knowledge and the literature. However, the list of such annotations is incomplete (only 20% of Pfam domains and families possess MF GO functional annotation). We therefore developed the GODomainMiner approach to expand the available functional annotations of protein domains and families. Based on our ECDomainMiner approach, we use the respective associations of protein sequences with GO terms and protein domains to infer direct associations between GO terms and protein domains. Finally, we used our calculated GO-Domain associations to devise a systematic way, called AutoProf-Annotator, to generate high confidence rules for protein sequence (or structure) annotation.