Overview
Genome annotation is the process of attaching biological meaning to the raw sequence of an assembled genome. It identifies the locations of genes, their exon-intron boundaries, regulatory sequences, repetitive elements, and non-coding RNAs. Annotation bridges the gap between a static DNA sequence and the dynamic biological functions it encodes. Both computational predictions and experimental evidence are integrated to produce a comprehensive map of genomic features. As genome sequencing becomes faster and cheaper, the annotation bottleneck, transforming sequence data into biological insight, has become increasingly critical.
Methods
Annotation strategies fall into three categories. Ab initio prediction uses statistical models of gene structure (such as hidden Markov models) to identify coding regions directly from sequence composition. Homology-based annotation aligns expressed sequence tags, proteins, or RNA-seq reads from the same or related species to infer gene structures. Comparative annotation leverages evolutionary conservation across multiple species to pinpoint functional elements. Pipelines such as the NCBI Eukaryotic Genome Annotation Pipeline combine all three approaches, followed by manual curation to resolve ambiguous cases. Quality is assessed through metrics like the Annotation Edit Distance (AED).
Applications
Accurate annotation is essential for interpreting sequencing projects. In biomedical research, it enables the discovery of disease-causing mutations by revealing which genomic regions encode proteins or regulatory elements. Agricultural genomics uses annotation to link genes with traits such as yield and stress tolerance. Techniques like recombinant DNA technology depend on reliable gene models for cloning and expression. Annotation also supports bacterial genetics by identifying operons and virulence factors, while studies of gene regulation and epigenetics rely on the precise coordinates of promoters, enhancers, and other regulatory features.
Practical Protocol
The MAKER annotation pipeline integrates ab initio predictions, protein and transcript evidence into a unified annotation. Prepare a maker_opts.ctl file specifying the genome FASTA, repeat library (e.g., from RepeatModeler), and EST/protein alignments from closely related organisms. Run RepeatMasker to soft-mask repeats: RepeatMasker -species mammal genome.fa. Execute MAKER in three iterative rounds: first with only EST and protein evidence (maker -g genome.fa), which generates initial gene models via spliced alignments of transcripts and proteins. In the second round, train ab initio gene predictors (AUGUSTUS or SNAP) on high-confidence MAKER models from round one: snap -train -c score 1 -gff maker.gff genome.fa. The third round reruns MAKER with the trained predictors enabled, producing the final annotation. BRAKER2 extends this approach for species with RNA-seq data: align reads with STAR, run BRAKER2.pl --genome=genome.fa --bam=aligned.bam --species=species, which automatically trains GeneMark-ET and AUGUSTUS. Annotation quality is assessed with the Annotation Edit Distance (AED) metric: AED < 0.25 indicates high-confidence models. BUSCO completeness against the appropriate lineage is expected above 90% for a well-annotated genome. Compare your annotation to existing ones with gffcompare. For non-coding RNA annotation, run Infernal with Rfam covariance models. Functional annotation of predicted proteins uses InterProScan: interproscan.sh -i proteins.faa -f TSV -dp, which assigns Gene Ontology terms and domain annotations. The final GFF3 file should be sorted, validated with genometools gt gff3validator, and indexed with tabix.
