- Functional annotation using sequence similarity is a fundamental approach in bioinformatics for predicting the possible biological roles of genes and proteins. It involves comparing an uncharacterized DNA, RNA, or protein sequence with sequences that have already been studied and assigned functional information. If a new sequence shows significant similarity to a well-characterized sequence, researchers can use the known biological information to infer its possible function. This approach is widely used in genome annotation, protein characterization, comparative genomics, evolutionary biology, and functional genomics.
- Genome sequencing projects generate large numbers of predicted genes and protein sequences, but identifying their biological functions requires additional analysis. A newly predicted gene may have a complete coding sequence but no experimentally established function. Sequence similarity provides a way to investigate such genes by comparing them with reference sequences in biological databases. The results may suggest whether a protein is an enzyme, receptor, transporter, transcription factor, structural protein, or another type of biological molecule. They can also provide clues about conserved domains, molecular activities, cellular locations, and participation in biological pathways.
- The underlying principle of sequence similarity-based annotation is that biologically related sequences often retain recognizable similarities because they have evolved from a common ancestral sequence. When a gene or protein performs an important biological function, certain regions may remain conserved through evolution. These conserved regions can provide evidence of functional relationships between sequences from different organisms. However, sequence similarity does not automatically establish that two proteins perform exactly the same function. Functional inference must consider the strength and extent of similarity, evolutionary relationships, protein domains, sequence coverage, and biological context.
- Sequence similarity searches can be performed using nucleotide or protein sequences. Nucleotide comparisons are useful for identifying closely related genes, transcripts, and conserved genomic regions. Protein sequence comparisons are often more informative for functional annotation because the genetic code allows many nucleotide substitutions to occur without changing the encoded amino acid. Protein sequences also preserve important structural and functional features that may remain conserved across more distantly related organisms.
- BLAST (Basic Local Alignment Search Tool) is one of the most widely used programs for sequence similarity searching. It identifies local regions of similarity between a query sequence and sequences in a reference database. Different BLAST programs are designed for different types of comparisons. BLASTP compares a protein query against a protein database, while BLASTN compares a nucleotide query against a nucleotide database. BLASTX translates a nucleotide query into protein sequences in six reading frames and searches a protein database. TBLASTN compares a protein query against a nucleotide database translated in six reading frames, while TBLASTX translates both query and database nucleotide sequences before comparison.
- For protein functional annotation, BLASTP is commonly used to compare predicted protein sequences against databases such as UniProtKB, NCBI RefSeq Protein, or other curated and specialized protein resources. A significant match may identify a homologous protein with an established function. The annotation can then be transferred cautiously to the query sequence, particularly when the match covers most of the protein and involves a well-characterized reference. However, a match to an uncharacterized protein or a broadly annotated protein provides limited functional information.
- An illustrative BLASTP command is shown below:
# Compare predicted proteins against a protein database
blastp -query predicted_proteins.faa \
-db reference_proteins \
-out blast_results.tsv \
-evalue 1e-5 \
-outfmt 6 \
-num_threads 8
- The query file contains predicted protein sequences, while the database contains reference proteins. The E-value threshold controls which matches are reported, although the appropriate threshold depends on the purpose of the analysis, database size, sequence characteristics, and desired sensitivity. The tabular output can be used for downstream annotation and filtering. In a real workflow, researchers should record the database release, search parameters, and software version to ensure reproducibility.
- Several statistical measures are important when interpreting sequence similarity results. The E-value estimates the number of matches with a score at least as strong as the observed match that would be expected by chance in a database search under the statistical model. Lower E-values generally indicate stronger statistical evidence against a chance match. However, an E-value is influenced by database size and search conditions, so it should not be interpreted as the probability that a predicted function is correct.
- The bit score provides a normalized measure of alignment strength. Higher bit scores indicate stronger alignments. Percentage identity represents the proportion of aligned positions containing identical residues or nucleotides, while percentage similarity in protein alignments may also consider conservative amino acid substitutions. Query coverage indicates how much of the query sequence is included in the alignment, whereas subject coverage describes the proportion of the reference sequence covered. These measures are complementary: a high percentage identity over a very short region may provide weaker evidence for full-protein functional equivalence than a substantial alignment covering most of both proteins.
- Alignment length and biological context are also important. A query protein that matches only a short conserved domain may share a particular structural or biochemical feature with the reference protein without performing the same overall function. For example, several unrelated protein families may contain similar nucleotide-binding or catalytic domains. A full-length match with strong sequence similarity, conserved domain organization, and compatible biological context generally provides stronger evidence for functional transfer than a short local match.
- Homology is an evolutionary concept describing sequences that share common ancestry. Similarity is an observable property of sequences, while homology is an inference about their evolutionary relationship. Two sequences are either homologous or not; it is not scientifically precise to describe them as having a particular percentage of homology. Instead, researchers report percentage identity or similarity and use the evidence to assess whether the sequences are likely homologous.
- Orthologs and paralogs are particularly important when transferring functional annotations. Orthologs are genes in different species that evolved from a common ancestral gene through speciation. They often retain similar biological functions, making them useful for functional annotation across organisms. Paralogs arise through gene duplication and may evolve different functions or expression patterns. A sequence similarity search can identify both orthologs and paralogs, so the strongest sequence match is not necessarily the correct functional equivalent. Orthology inference may require phylogenetic analysis, gene-family comparisons, conserved synteny, and other evolutionary evidence.
- Reciprocal best-hit analysis is a simple approach sometimes used to identify potential orthologs. A query protein is first searched against a reference proteome, and the best-matching reference protein is then searched back against the original proteome. If the original query is recovered as the best match, the pair may be considered a candidate orthologous relationship. However, reciprocal best hits do not reliably resolve every gene family, particularly when gene duplications, gene losses, unequal evolutionary rates, or incomplete genome annotations are involved. More comprehensive orthology tools such as OrthoFinder, eggNOG, and OMA can provide additional evolutionary information.
- Functional annotation using sequence similarity commonly relies on several biological databases. UniProtKB is a major resource for protein sequence and functional information. Its reviewed Swiss-Prot section contains manually curated protein records, while the unreviewed TrEMBL section contains computationally annotated records. NCBI RefSeq provides reference nucleotide and protein sequences with associated annotation. The NCBI nonredundant protein database contains sequences from multiple sources and is widely used for similarity searches, although its records differ in evidence quality and annotation depth.
- Other resources provide specialized information. The Protein Information Resource and related protein resources support protein classification and annotation. InterPro integrates information from multiple protein family, domain, and functional signature databases. Pfam provides protein families and profile hidden Markov models, while PROSITE identifies conserved patterns and profiles associated with protein families and functional sites. The Conserved Domain Database (CDD) provides information about conserved protein domains. These resources complement sequence similarity searches by identifying specific structural and functional features within proteins.
- A common functional annotation workflow begins with predicted protein sequences in FASTA format. The sequences are compared against a reference protein database using BLASTP, DIAMOND, or another suitable search tool. Significant matches are filtered according to alignment quality, coverage, identity, and statistical evidence. The best-supported matches are then examined for protein names, gene names, molecular functions, domains, enzyme activities, and other annotations. Additional analysis may be performed using InterProScan, orthology databases, Gene Ontology, and pathway resources.
- DIAMOND is a high-performance sequence alignment program designed for rapid protein similarity searches. It is particularly useful for large genome and metagenome datasets, where conventional BLAST searches may require substantial computational time. DIAMOND supports protein-to-protein searches and translated nucleotide-to-protein searches. An illustrative command is:
# Build a DIAMOND protein database
diamond makedb –in reference_proteins.faa \
-d reference_db
# Search predicted proteins against the database
diamond blastp –query predicted_proteins.faa \
–db reference_db \
–out diamond_results.tsv \
–evalue 1e-5 \
–outfmt 6 \
–threads 8
- DIAMOND results can be used to identify candidate homologous proteins and transfer functional information from suitable reference records. The database should be selected according to the organism and research objective, and its release should be documented. Rapid searches do not eliminate the need for careful biological interpretation.
- A tabular search output can contain query identifiers, reference identifiers, percentage identity, alignment length, mismatch count, gap count, query and subject coordinates, E-value, and bit score. For example, a simplified BLAST tabular output may look like this:
query_id subject_id pident length qcov evalue bitscore
protein1 ref_protein1 85.2 320 98 1e-120 410
protein2 ref_protein2 42.7 180 55 2e-25 105
protein3 ref_protein3 31.5 75 20 1e-04 45
- The values are illustrative and do not represent actual search results. The first match provides stronger evidence for a close relationship because it has high identity and extensive query coverage. The second may indicate a more distant relationship or a partial match. The third requires particular caution because the alignment covers only a small part of the query. Researchers should also examine subject coverage and determine whether the aligned region corresponds to a conserved domain or a complete protein.
- Functional annotation can be improved by identifying conserved protein domains and motifs. A domain is a distinct structural or functional unit within a protein that may contribute to activities such as ligand binding, catalysis, DNA binding, or protein–protein interaction. Many proteins contain multiple domains, and the combination and order of these domains can provide important clues about their biological roles. Domain analysis can therefore reveal information that may not be apparent from a single full-length similarity match.
- InterProScan is widely used to identify protein signatures by searching multiple integrated member databases. It can provide information about protein families, domains, conserved sites, and associated Gene Ontology annotations. A simplified command-line example is:
# Search protein sequences using InterProScan
interproscan.sh -i predicted_proteins.faa \
-f TSV,GFF3 \
-goterms \
-iprlookup \
-o interpro_results.tsv
- InterProScan output can be used to identify conserved domains and protein-family signatures. The available applications and output options depend on the installed version and configuration. Researchers should consult the relevant documentation and retain the InterProScan release information when reporting results.
- Gene Ontology (GO) provides a standardized framework for describing gene and protein functions. It organizes annotations into three main categories: molecular function, biological process, and cellular component. Molecular function describes the biochemical activities performed by a gene product, such as ATP binding, DNA binding, or catalytic activity. Biological process describes broader biological objectives involving coordinated activities, such as DNA replication, cell division, or carbohydrate metabolism. Cellular component identifies the locations where gene products are found, such as the nucleus, mitochondrion, membrane, or ribosome.
- Sequence similarity can be used to transfer GO annotations from characterized proteins to uncharacterized proteins. However, annotation transfer should consider the evidence supporting the reference annotation and the evolutionary relationship between the sequences. A protein that is similar to a known enzyme may share its catalytic activity but differ in substrate specificity or cellular role. GO annotations should therefore be assigned with suitable evidence codes and appropriate levels of specificity. Computational predictions should not be presented as experimentally confirmed functions.
- Enzyme annotation is another important application of sequence similarity. Enzymes can be classified using Enzyme Commission (EC) numbers, which describe the reactions they catalyse. A protein that resembles a known enzyme may be assigned a candidate EC classification, but the complete catalytic activity should be supported by conserved active-site residues, domain architecture, and appropriate sequence coverage. Closely related enzymes can catalyse different reactions, and some proteins may have multiple activities. Sequence similarity alone may therefore be insufficient to establish a precise substrate or reaction.
- Pathway annotation uses predicted protein functions to identify the biological pathways in which genes may participate. Databases such as KEGG and Reactome provide information about biochemical reactions, molecular interactions, and biological pathways. When a predicted protein is assigned a function through sequence similarity or domain analysis, it may be mapped to a corresponding pathway. Pathway assignments are more reliable when several components of the same pathway are identified and the predicted functions are supported by independent evidence. A single similarity match should not be treated as proof that an organism possesses an entire pathway.
- Functional annotation is also important in metagenomics. Metagenomic sequencing produces DNA sequences from microbial communities, and gene prediction identifies candidate coding regions within assembled contigs or directly from sequencing reads. Protein sequences are then compared against reference databases to identify likely functions. Tools such as DIAMOND and databases such as eggNOG, UniProt, and KEGG can support this process. Functional profiling can reveal the metabolic potential of microbial communities, including pathways related to carbon cycling, nitrogen metabolism, nutrient utilization, and environmental adaptation. However, detecting a gene indicates potential biological capability rather than proving that the corresponding process is active under the sampled conditions.
- A major challenge in functional annotation is the presence of hypothetical and uncharacterized proteins. These proteins may have recognizable sequence similarity to other hypothetical proteins but lack experimentally established functions. In such cases, assigning a precise biological role is not justified. They may be described as conserved hypothetical proteins, uncharacterized proteins, or proteins containing a particular conserved domain, depending on the available evidence. As new experimental and computational information becomes available, their annotations may be refined.
- Another limitation is the propagation of annotation errors. If an incorrectly annotated protein is used as a reference for new sequence annotations, its inaccurate description may be transferred to many additional records. This can create chains of incorrect functional assignments in biological databases. The problem is especially important in large-scale automated annotation projects, where thousands or millions of sequences may be annotated through similarity-based methods. Using curated reference proteins, checking evidence codes, and distinguishing experimental annotations from computational predictions can reduce this risk.
- Functional divergence after gene duplication is another important source of uncertainty. Paralogs may retain similar sequences while acquiring different biochemical activities, substrate preferences, expression patterns, or cellular locations. Even orthologous proteins may undergo functional changes during evolution. Therefore, annotation transfer should be supported by evolutionary analysis, conserved domain organization, catalytic residues, gene context, and experimental information whenever possible.
- Sequence similarity searches can also produce false positives and false negatives. False positives may arise from short conserved motifs, low-complexity regions, repetitive sequences, or chance similarities. False negatives may occur when sequences are highly divergent, incomplete, or poorly represented in reference databases. Low-complexity masking, suitable scoring matrices, profile-based searches, and domain analysis can improve sensitivity and specificity. Profile hidden Markov models, such as those used by HMMER and Pfam, are particularly useful for detecting distant relationships that may be missed by direct pairwise sequence comparisons.
- Functional annotation should therefore integrate multiple evidence sources rather than rely exclusively on the highest-scoring sequence match. Useful supporting information includes conserved domains, protein motifs, predicted structure, subcellular localization, orthology relationships, gene neighbourhood, expression patterns, protein–protein interactions, and experimental evidence. Structural prediction and comparison can sometimes reveal similarities that are difficult to detect at the sequence level. Integration of these methods provides a more reliable and biologically meaningful functional description.
- For reproducible functional annotation, researchers should record the input sequence identifiers, reference database names and releases, search software and versions, scoring parameters, E-value thresholds, filtering criteria, and annotation-transfer rules. They should retain raw search results and distinguish direct experimental evidence from computational inference. Accession numbers and database versions should be preserved so that annotations can be traced to their original records. This is especially important because protein databases are continually updated and functional descriptions may change as new evidence becomes available.
- Functional annotation using sequence similarity is a powerful approach for assigning possible biological roles to newly identified genes and proteins. It connects unknown sequences with existing biological knowledge and provides a foundation for protein characterization, pathway analysis, comparative genomics, and functional genomics. However, reliable annotation requires more than selecting the best database hit. Researchers must evaluate statistical significance, sequence coverage, evolutionary relationships, domain architecture, functional evidence, and biological context. Combining sequence similarity with domain analysis, orthology inference, Gene Ontology, pathway databases, and experimental validation produces more accurate and reproducible functional annotations.