- Genome annotation is the process of identifying biological features within a genome and assigning information about their structure and biological roles. It transforms a raw genome sequence into an organized representation of genes, transcripts, regulatory regions, noncoding RNAs, repetitive elements, and other genomic components. Two major components of this process are structural genome annotation and functional genome annotation. Structural annotation identifies the locations, boundaries, and organization of genomic features, whereas functional annotation determines or predicts what these features do biologically. Together, they provide the foundation for understanding genome organization, gene expression, protein function, biological pathways, genetic variation, evolution, and disease-associated mechanisms.
- Structural genome annotation focuses on identifying the physical organization of genes and other functional elements within a genome. It determines where a feature begins and ends, which strand it occupies, how its component regions are arranged, and how different features are related. For protein-coding genes, structural annotation identifies gene boundaries, transcription start sites, exons, introns, coding sequences, untranslated regions, and transcription termination regions. For noncoding genes, it identifies the genomic positions and structures of RNA genes such as transfer RNA, ribosomal RNA, microRNA, small nuclear RNA, and long noncoding RNA genes. It may also identify pseudogenes, promoters, repetitive elements, transposable elements, and other genomic features.
- The first step in structural annotation is generally the preparation of a high-quality genome assembly. Genome assembly reconstructs genomic sequences from sequencing reads and organizes them into contigs, scaffolds, or chromosome-level sequences. Assembly quality strongly influences annotation accuracy. Fragmented assemblies can interrupt genes, separate exons across contigs, or produce incomplete gene models. Misassemblies may create false gene structures or incorrectly join genomic regions. Repetitive sequences can also complicate assembly and gene identification. Therefore, assembly completeness, contamination, sequence accuracy, and chromosome organization should be assessed before annotation begins.
- Gene prediction is a central component of structural annotation. It involves identifying genomic regions that are likely to represent genes and determining their boundaries and internal organization. Protein-coding gene prediction is particularly challenging in eukaryotes because genes commonly contain exons separated by introns. A single gene may produce multiple transcript isoforms through alternative promoters, alternative splicing, and alternative polyadenylation. Consequently, structural annotation may need to identify several transcript models for one gene rather than assigning only one continuous coding region.
- In prokaryotes, gene prediction is generally more straightforward because bacterial and archaeal genes often have compact structures with continuous coding sequences. Many prokaryotic genes occur in operons, where several genes are transcribed together from a shared promoter. Gene prediction methods identify likely start codons, stop codons, coding regions, ribosome-binding sites, and other characteristic sequence patterns. Tools such as Prodigal, GeneMark, and Glimmer have been widely used for microbial gene prediction. The identification of operons and transcriptional organization may require additional analysis beyond the prediction of individual coding sequences.
- In eukaryotes, structural annotation must identify the exon–intron organization of genes. Exons are retained in mature RNA transcripts, while introns are generally removed through RNA splicing. Splice-site signals, such as the common GT–AG intron boundaries in DNA notation, provide evidence for intron positions, although other splice-site patterns also occur. Computational gene predictors use these signals together with coding potential, codon usage, nucleotide composition, and statistical models to predict gene structures. Ab initio programs such as AUGUSTUS and GeneMark-ES use probabilistic models to identify likely genes, while other methods incorporate transcript and protein evidence.
- Structural annotation can use three broad types of evidence: ab initio predictions, homology-based evidence, and experimental evidence. Ab initio prediction identifies genes using sequence characteristics without requiring direct comparison to known genes. Homology-based methods compare the genome with known nucleotide or protein sequences from the same or related organisms. Experimental evidence includes RNA-seq reads, full-length transcript sequences, expressed sequence tags, cDNA sequences, and proteomic data. Combining these evidence types generally improves gene-model reliability because each provides different information about gene structure.
- RNA sequencing is particularly useful for identifying expressed genes and supporting transcript structures. RNA-seq reads can be aligned to the genome to identify transcribed regions, exon boundaries, splice junctions, and alternative transcript isoforms. Full-length transcript sequencing using long-read technologies can provide more complete evidence for transcript structures, especially when several exons must be connected into a single isoform. Protein sequences from related organisms can also support gene prediction by identifying conserved coding regions. However, transcript and protein evidence may be incomplete, tissue-specific, or derived from related but evolutionarily different species. Such evidence must therefore be interpreted in the context of the organism being annotated.
- Structural annotation also identifies untranslated regions, which are parts of mature messenger RNA that are not translated into proteins. The 5′ untranslated region occurs before the coding sequence, while the 3′ untranslated region follows it. These regions can influence translation efficiency, RNA stability, intracellular localization, and post-transcriptional regulation. Their boundaries may be difficult to determine accurately without transcript evidence, particularly when alternative transcription start sites or polyadenylation sites are present.
- Noncoding RNA annotation is another important component of structural genome annotation. Transfer RNA genes encode molecules that carry amino acids during protein synthesis, while ribosomal RNA genes encode structural and catalytic components of ribosomes. MicroRNAs regulate gene expression by interacting with target messenger RNAs. Small nuclear RNAs participate in RNA processing, and small nucleolar RNAs contribute to RNA modification and ribosome biogenesis. Long noncoding RNAs have diverse regulatory and structural functions. Tools such as tRNAscan-SE identify transfer RNA genes, while Infernal uses covariance models to detect structured RNA families. Rfam provides curated RNA family information and models that support the identification of known noncoding RNA families.
- Other genomic elements are also identified during structural annotation. Pseudogenes are sequences that resemble functional genes but have generally lost the ability to produce the original functional product. They may arise through gene duplication followed by loss of function or through retrotransposition. Repetitive DNA includes tandem repeats, microsatellites, satellite DNA, and interspersed transposable elements. Transposable elements include DNA transposons, long interspersed nuclear elements, short interspersed nuclear elements, and long terminal repeat retrotransposons. These elements can affect genome structure, gene regulation, and genome evolution. RepeatMasker and RepeatModeler are commonly used to identify and characterize repetitive sequences.
- Regulatory elements are another important category of genomic features. Promoters are regions associated with transcription initiation, while enhancers and silencers can increase or decrease gene expression. Insulators and other regulatory elements can influence interactions between genomic regions. Transcription factor binding sites are short DNA sequences recognized by regulatory proteins. Their identification may use sequence motif searches, comparative conservation, chromatin accessibility data, and experimental evidence from techniques such as ChIP-seq. Regulatory annotation is challenging because regulatory activity can vary across tissues, developmental stages, and environmental conditions.
- Functional genome annotation focuses on assigning biological meaning to genes and their products. Once structural annotation has identified a gene or transcript, functional annotation attempts to determine its likely role in the organism. For protein-coding genes, this may include predicting the protein product, molecular activity, biological process, cellular location, conserved domains, enzyme classification, and pathway involvement. For noncoding RNA genes, functional annotation may describe their RNA family, molecular interactions, regulatory roles, or involvement in RNA processing. Functional annotation can be based on experimental evidence, sequence similarity, conserved protein domains, evolutionary relationships, and computational predictions.
- Sequence similarity searching is one of the most widely used approaches to functional annotation. Predicted protein sequences are compared against databases containing previously characterized proteins. BLASTP, DIAMOND, and related tools identify similar sequences that may provide evidence about the function of an unknown protein. A strong match to a well-characterized protein can support a functional assignment, particularly when sequence coverage and conserved residues are also consistent. However, sequence similarity does not always establish identical biological function. Proteins within the same family may have different substrates, activities, cellular locations, or biological roles. Functional assignments should therefore consider the strength and biological relevance of the evidence rather than relying on similarity scores alone.
- Protein domain and motif analysis provides additional information about protein function. Many proteins contain conserved domains that are associated with specific biochemical activities, structural properties, or molecular interactions. A protein may contain one domain or several domains arranged in a particular order. Domain composition can provide clues about its function even when its overall sequence similarity to a known protein is limited. InterPro integrates information from multiple protein signature databases, including Pfam, PROSITE, SMART, and related resources. HMMER can search protein sequences against profile hidden Markov models representing conserved protein families and domains. The NCBI Conserved Domain Database provides additional information about conserved protein domains and their functional characteristics.
- Gene Ontology is a widely used framework for describing gene and protein functions in a standardized manner. It organizes functional information into three major categories: molecular function, biological process, and cellular component. Molecular function describes the biochemical activity of a gene product, such as DNA binding, ATP binding, or catalytic activity. Biological process describes the broader biological objective in which the gene product participates, such as DNA replication, cell division, metabolism, immune response, or development. Cellular component describes the location where the gene product acts, such as the nucleus, mitochondrion, ribosome, cytoplasm, or plasma membrane. Gene Ontology annotations enable functional comparisons across organisms and support gene-set enrichment analysis.
- Pathway annotation connects genes and proteins to broader biological and biochemical processes. KEGG and Reactome are important resources for identifying metabolic pathways, signaling pathways, molecular interactions, and biological processes. Enzyme classification can be supported by the Enzyme Commission system, which organizes enzymes according to the reactions they catalyse. Pathway information helps researchers understand how individual genes contribute to cellular activities and biological systems. However, pathway assignments may be incomplete for poorly studied organisms, and predicted pathway involvement should not be confused with experimentally demonstrated activity.
- Functional annotation also involves predicting protein localization, membrane association, secretion signals, and structural characteristics. Subcellular localization prediction attempts to determine whether a protein is likely to function in the nucleus, cytoplasm, mitochondria, chloroplasts, endoplasmic reticulum, or other cellular compartments. Signal peptide prediction can identify proteins likely to enter secretory pathways, while transmembrane helix prediction can identify membrane-associated proteins. These predictions help characterize the possible biological roles of proteins, but they should be treated as computational evidence rather than definitive experimental confirmation.
- Enzyme annotation is particularly important for understanding metabolism. Enzymes catalyse biochemical reactions and are commonly classified according to the reactions they perform. Functional annotation may identify enzyme families, catalytic residues, cofactor requirements, and possible substrates. Databases such as KEGG, MetaCyc, and BRENDA provide information about metabolic reactions and enzyme functions. Nevertheless, assigning a specific enzyme activity from sequence similarity alone can be difficult, especially when closely related enzymes act on different substrates or catalyse related reactions.
- Functional annotation also includes the identification of gene families and evolutionary relationships. Orthologous genes are genes in different species that descended from a common ancestral gene through speciation, while paralogous genes arise through gene duplication. Orthology can support functional inference because orthologous genes often retain related biological functions, although functional divergence can occur. Comparative resources such as Ensembl, OrthoDB, and eggNOG provide information about orthologous groups and gene families. Such information is useful for transferring functional knowledge from well-studied organisms to less-characterized species, but transferred annotations should be assigned appropriate confidence levels.
- Several databases support structural and functional genome annotation. GenBank, the European Nucleotide Archive, and the DNA Data Bank of Japan provide nucleotide sequence records and associated annotations. RefSeq offers curated reference sequences for many organisms. Ensembl and Ensembl Genomes provide genome assemblies, gene models, transcript information, comparative genomics, and functional annotation. UniProt provides protein sequence and functional information, while InterPro supports protein family and domain annotation. Gene Ontology, KEGG, Reactome, and related resources provide standardized functional and pathway information. Organism-specific databases, including FlyBase, WormBase, TAIR, and the Saccharomyces Genome Database, offer detailed information for particular model organisms.
- Structural and functional annotation are often performed through integrated pipelines. For prokaryotic genomes, a pipeline may identify coding sequences, transfer RNA genes, ribosomal RNA genes, other RNA features, and repetitive regions before assigning functions to predicted proteins. Prokka is a widely used rapid annotation pipeline for bacterial and archaeal genomes. For eukaryotic genomes, workflows may combine repeat identification, gene prediction, RNA-seq alignment, protein homology, transcript assembly, noncoding RNA detection, and functional annotation. BRAKER and MAKER are examples of systems used in eukaryotic genome annotation workflows. The choice of pipeline depends on the organism, genome assembly, available experimental evidence, and the required level of annotation detail.
- The following example demonstrates a simplified structural annotation workflow for a bacterial genome using Prokka. The input genome should be a quality-controlled assembly in FASTA format.
# Annotate a bacterial genome
prokka \
–outdir prokka_output \
–prefix sample \
–kingdom Bacteria \
genome.fasta
- Prokka generates several annotation files, including GFF, GenBank, nucleotide FASTA, protein FASTA, and summary files. Its functional assignments depend on the available databases and software version. Prokka is designed for prokaryotic genome annotation and is not a general-purpose eukaryotic annotation pipeline.
- For eukaryotic genome annotation, BRAKER can integrate protein evidence and, depending on the workflow, RNA-seq evidence to support gene prediction. The following illustrates an invocation using protein evidence; actual requirements and options should be checked against the installed BRAKER version.
# Example BRAKER workflow using protein evidence
braker.pl \
–genome=genome.fasta \
–prot_seq=proteins.fasta \
–species=sample_species \
–workingdir=braker_output
- Functional annotation can be performed on the predicted protein sequences using InterProScan. The following example searches protein sequences against supported signature databases and requests Gene Ontology information where available.
# Functional annotation using InterProScan
interproscan.sh \
-i proteins.fasta \
-f tsv \
-o interpro_results.tsv \
-goterms \
-iprlookup
- The resulting file may contain protein identifiers, conserved domains, signature matches, InterPro accessions, and associated Gene Ontology terms. The results provide evidence for possible protein functions, but not every predicted protein will receive a reliable functional assignment.
- Structural and functional annotation results are stored in several standard bioinformatics formats. GFF3 and GTF describe genomic features and their relationships, including genes, transcripts, exons, and coding sequences. GenBank flat files combine sequence information with feature annotations, while FASTA files store nucleotide or protein sequences. Tabular formats are frequently used to report functional annotations, database matches, domain predictions, and Gene Ontology terms. Consistency between genome assemblies, chromosome identifiers, gene identifiers, transcript identifiers, and coordinate systems is essential when combining these files.
- The following simplified GFF3 example illustrates how structural annotation represents a gene and its transcript components:
##gff-version 3
chr1 annotation gene 1000 2500 . + . ID=gene001
chr1 annotation mRNA 1000 2500 . + . ID=transcript001;Parent=gene001
chr1 annotation exon 1000 1200 . + . ID=exon001;Parent=transcript001
chr1 annotation exon 1800 2500 . + . ID=exon002;Parent=transcript001
chr1 annotation CDS 1050 1200 . + 0 ID=cds001;Parent=transcript001
chr1 annotation CDS 1800 2400 . + 1 ID=cds002;Parent=transcript001
- This example illustrates the hierarchical relationship between a gene, its transcript, exons, and coding sequence segments. A complete GFF3 file must follow the format specification, including valid feature relationships, coordinate conventions, attribute encoding, and correct CDS phase values. Structural annotation should also represent alternative transcripts separately when a gene produces multiple transcript isoforms.
- Quality assessment is essential for both structural and functional genome annotation. Structural annotation can be evaluated by examining gene completeness, exon–intron structures, transcript support, coding sequence integrity, and agreement with known proteins. BUSCO is commonly used to assess the completeness of genome assemblies and predicted gene sets by identifying expected conserved single-copy orthologs for a selected lineage. It reports complete, duplicated, fragmented, and missing BUSCO groups. However, BUSCO completeness does not establish that every gene model is correct, and its results depend on the selected lineage dataset and the evolutionary characteristics of the organism.
- Additional structural validation can use RNA-seq read support, full-length transcript evidence, protein alignments, conserved splice sites, and comparison with annotations from closely related species. Genome browsers such as IGV and JBrowse allow researchers to inspect gene models alongside sequencing reads and other supporting evidence. Tools such as AGAT can assist with the validation, conversion, and manipulation of annotation files. These methods help identify incorrect exon boundaries, incomplete transcripts, unsupported gene predictions, and inconsistencies between annotation files.
- Functional annotation quality is more difficult to assess because many predicted proteins have no experimentally characterized homologues. A predicted protein may contain a recognizable domain but still have an uncertain specific function. Automated annotation systems may transfer descriptions from related proteins, potentially propagating errors when the original annotation is incorrect or too general. Functional evidence should therefore be classified according to its source, such as direct experimental evidence, curated database information, sequence similarity, conserved domain evidence, or computational prediction. Terms such as hypothetical protein or protein of unknown function are appropriate when reliable functional evidence is unavailable.
- Structural and functional annotation face different but related challenges. Structural annotation can be affected by fragmented genome assemblies, sequencing errors, repetitive regions, alternative splicing, incomplete transcript evidence, and incorrect gene prediction parameters. Functional annotation can be limited by incomplete reference databases, evolutionary divergence, poorly characterized protein families, and uncertain functional transfer. Both are affected by the availability and quality of biological evidence. A well-predicted gene may still have an unknown function, while a protein with a known function may be difficult to annotate correctly if its gene structure has been predicted inaccurately.
- The distinction between structural and functional annotation is particularly important in genome research. Structural annotation answers questions such as where a gene is located, how many exons it contains, which transcript isoforms it produces, and where its coding sequence begins and ends. Functional annotation addresses questions such as what the gene product does, which biological processes it participates in, where it acts in the cell, and which pathways it may influence. Structural information provides the framework for functional interpretation, while functional evidence can sometimes help refine structural predictions. The two processes therefore interact and are often improved through iterative analysis.
- These annotation approaches have applications across medicine, agriculture, biotechnology, microbiology, and evolutionary biology. In medical genomics, structural annotation identifies genes and transcripts that may be affected by disease-associated variants, while functional annotation helps interpret their potential effects on proteins, pathways, and biological processes. In agriculture, annotation supports the discovery of genes associated with crop yield, stress tolerance, disease resistance, animal production, fertility, and adaptation. In microbial genomics, it helps identify metabolic capabilities, antibiotic resistance determinants, virulence-associated genes, and pathways involved in environmental adaptation. In evolutionary research, it supports the investigation of gene gain and loss, gene duplication, genome organization, and functional divergence.
- Genome annotation is not a one-time activity. New sequencing technologies, improved genome assemblies, additional transcriptomic and proteomic evidence, and updated databases can reveal previously unrecognized genes or correct existing annotations. Long-read sequencing can improve transcript reconstruction and help resolve complex gene structures. Proteomics can provide evidence that predicted proteins are expressed, while ribosome profiling can support the identification of translated regions. Functional experiments can confirm or refine computational predictions. Updated annotation releases may therefore contain revised gene boundaries, new transcript isoforms, improved protein descriptions, and additional functional information.
- For reproducible research, annotation workflows should record the genome assembly accession and version, annotation pipeline, software versions, database releases, input evidence, parameters, and filtering criteria. Researchers should preserve the original genome assembly and annotation outputs, maintain consistent gene and transcript identifiers, and document the evidence supporting functional assignments. When annotations are updated, changes should be recorded so that downstream analyses can be interpreted in relation to the correct annotation release. These practices are especially important when comparing studies or analysing genetic variants across different genome versions.
- Structural and functional genome annotation are complementary processes that convert genomic sequence data into biologically meaningful information. Structural annotation identifies the locations and organization of genes, transcripts, regulatory elements, and other genomic features, while functional annotation predicts or describes their molecular activities, biological processes, cellular locations, and pathway roles. Combining computational prediction with sequence similarity, transcriptomic evidence, protein information, conserved domains, ontologies, and experimental validation produces more informative and reliable genome annotations. Understanding both approaches is essential for interpreting genome sequences and conducting accurate research in modern bioinformatics.