Genome annotation

Genome annotation turns a raw string of A, C, G and T into a labeled map. It follows sequencing and assembly and has two broad stages. Structural annotation finds features and their exact positions: protein-coding genes with their exons and introns, genes for RNAs such as transfer and ribosomal RNA, pseudogenes, regulatory regions, repeats and transposons. Functional annotation then assigns likely roles, such as a gene name, a protein's enzyme activity or the pathway it belongs to.

Two kinds of evidence are combined. Ab initio methods predict genes from the sequence alone, using signals such as start and stop codons, splice sites and the statistical patterns typical of coding DNA. Homology-based methods align known proteins, transcripts from RNA sequencing and related genomes to the new sequence, since similarity to a well-studied gene suggests a shared function. Repeats are usually masked first so they do not produce false gene calls. Automated pipelines, such as NCBI's pipeline for bacterial and archaeal genomes, chain these steps together.

Annotation is never final. Most functions are inferred rather than tested in the lab, so one error can spread to many records, and genomes are reannotated as evidence improves. Annotations are stored in databases such as GenBank and viewed in genome browsers, and they underpin expression studies, the interpretation of disease-causing variants and comparisons between species.