A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Functional annotation assigns biological meaning to predicted genes and proteins — what do they do, where do they act, and what processes do they participate in? The primary approach is homology-based: BLAST or HMM searches against curated databases (UniProt, Pfam, InterPro) identify conserved domains and assign functions based on characterized homologs. Gene Ontology (GO) provides a standardized vocabulary for describing function across three axes: biological process, molecular function, and cellular component. Annotation pipelines combine sequence homology, domain architecture, orthology assignment, and genomic context to produce comprehensive functional predictions.
Take 10 uncharacterized protein sequences, run them through InterProScan, and examine the domain annotations. Then search each against UniProt/SwissProt with BLAST. Compare the information gained from domain architecture versus homology to well-studied proteins. Note cases where domain annotation succeeds but BLAST fails, and vice versa.
A genome assembly with gene predictions tells you where the genes are, but not what they do. Functional annotation is the process of determining the biological role of each gene product. For model organisms with decades of experimental study, many genes have experimentally determined functions. For newly sequenced organisms, computational prediction from sequence similarity is the primary route to functional understanding.
The core logic is homology-based transfer: if two proteins share significant sequence similarity (implying common ancestry), they likely share similar functions. In practice, this is implemented at two levels. Sequence-level searches (BLAST against UniProt/SwissProt) find the closest characterized relatives and transfer their annotations. Domain-level searches (InterProScan, which integrates Pfam, PROSITE, SMART, and other domain databases) identify conserved functional domains within the protein. A protein might not have a close full-length homolog in the database, but its individual domains — a kinase domain here, a DNA-binding domain there — can be recognized, providing modular functional information. Domain architecture (the specific combination and order of domains) is often more informative than overall sequence similarity for predicting function.
Gene Ontology (GO) provides the standardized vocabulary for functional annotation. Every GO term belongs to one of three hierarchies: biological process (what the gene does at the cellular or organismal level, e.g., "inflammatory response"), molecular function (what the gene product does biochemically, e.g., "protein kinase activity"), and cellular component (where the gene product acts, e.g., "nucleus"). GO terms are connected in a directed acyclic graph from general to specific, so annotating a gene with a specific term automatically implies all parent terms. GO annotation is the lingua franca of functional genomics — pathway enrichment analysis, functional comparison across species, and knowledge-based network analysis all depend on GO.
Orthology-based annotation leverages the principle that orthologs (genes separated by speciation) tend to conserve function more reliably than paralogs (genes separated by duplication). Tools like OrthoFinder, eggNOG, and OMA assign orthology groups across genomes and transfer functional annotations within ortholog groups. This is more reliable than simple best-BLAST-hit annotation because it accounts for gene family structure. Combined with synteny information (conserved genomic context) and expression data (conserved expression patterns), orthology-based annotation provides the most reliable computational functional predictions available. The ultimate goal — experimentally validating the function of every gene in every organism — remains distant, making computational annotation an essential and evolving discipline.