- E-values, bit scores, percentage identity, and query coverage are four important measures used to evaluate sequence similarity search results in bioinformatics. They are commonly reported by tools such as BLAST and FASTA and help researchers determine whether a database sequence is likely to be biologically related to a query sequence, how strong the alignment is, how closely the aligned residues match, and how much of the query sequence is represented in the alignment. Although these measures are related, each describes a different aspect of sequence comparison. Understanding their differences is essential for interpreting sequence search results accurately.
- An E-value, or expectation value, estimates the number of alignments with a score equal to or better than the observed score that would be expected to occur by chance when searching a database of a particular size. In sequence similarity searching, a lower E-value generally indicates that the observed alignment is less likely to have arisen randomly. For example, an E-value of 10 means that approximately ten alignments with an equivalent or better score might be expected by chance in a database search under the statistical model. An E-value of 0.01 indicates a much less frequent chance match, while an E-value of 1 × 10⁻⁵ suggests stronger statistical evidence against a random match.
- The E-value depends on several factors, including the alignment score, query length, database size, sequence composition, scoring system, and statistical parameters. A larger database increases the number of opportunities for chance matches, so the same alignment can receive a higher E-value when searched against a larger database. Consequently, E-values should be interpreted in the context of the database and search conditions rather than as universal measures of biological significance.
- In simplified form, the expected number of chance alignments can be expressed as:
[
E \approx Kmn e^{-\lambda S}
]
- Here, (E) is the E-value, (K) and (\lambda) are statistical parameters related to the scoring system and sequence composition, (m) and (n) represent effective query and database sequence lengths, and (S) is the raw alignment score. Actual search programs use more detailed statistical calculations, including adjustments for effective sequence lengths and search-space characteristics.
- An E-value is not the same as a probability that two sequences are unrelated. It represents an expected number of chance alignments, not the probability that a particular match is false. For small E-values, the probability of observing at least one such chance event can sometimes be approximated by (1-e^{-E}), under a suitable Poisson model. However, this approximation does not turn the E-value into a direct probability of biological homology. A statistically significant alignment may still lack biological relevance, especially when it involves low-complexity regions, repetitive sequences, biased amino acid composition, or a short conserved motif.
- A bit score is a normalized measure of alignment strength. It expresses the alignment score on a logarithmic scale and allows scores obtained using different search spaces or statistical settings to be compared more meaningfully than raw scores alone. Higher bit scores indicate stronger alignments. Unlike raw alignment scores, bit scores are normalized using statistical parameters associated with the scoring system.
- The relationship between a raw alignment score and a bit score is commonly expressed as:
[
S’ = \frac{\lambda S-\ln K}{\ln 2}
]
- Here, (S’) is the bit score, (S) is the raw alignment score, and (K) and (\lambda) are statistical parameters. A bit score is therefore not simply the raw score divided by a fixed number; its calculation depends on the statistical properties of the scoring system.
- Bit scores and E-values are closely related. For a fixed search space and statistical model, a higher bit score generally corresponds to a lower E-value. A useful approximate relationship is:
[
E \approx mn \times 2^{-S’}
]
- where (m) and (n) represent effective query and database lengths. This simplified expression illustrates why an increase in bit score can substantially reduce the expected number of chance matches. However, the exact relationship depends on the effective search space and the statistical model used by the alignment program.
- Percentage identity measures the proportion of aligned positions containing exactly matching residues. In nucleotide alignments, it represents the percentage of positions containing the same nucleotide, while in protein alignments it represents the percentage containing the same amino acid. A high percentage identity indicates that the aligned region is highly similar at the residue level, but it does not automatically establish that the sequences have the same biological function.
- Percentage identity is commonly calculated as:
[
\text{Percentage identity} =
\frac{\text{Number of identical aligned positions}}
{\text{Number of positions included in the identity calculation}}
\times 100
]
- The denominator must be interpreted carefully because software and reporting conventions can differ in how alignment gaps and ambiguous residues are handled. In many alignment reports, identity is calculated using the total number of alignment columns, including columns containing gaps. For example, if an alignment contains 100 columns and 85 positions have identical residues, the reported identity is 85%. Researchers should check the software documentation and output definition before comparing identity values from different tools.
- Identity is different from similarity, particularly in protein sequence analysis. Similarity may include conservative amino acid substitutions in which different residues have comparable chemical properties and receive positive scores under a substitution matrix such as BLOSUM62 or PAM. For example, leucine and isoleucine are different amino acids and therefore do not count as identical, but they may be considered similar because of their related chemical properties. Thus, a protein alignment can have a lower percentage identity but a higher percentage similarity when conservative substitutions are present.
- Query coverage indicates how much of the query sequence is included in an alignment with a database sequence. It is commonly expressed as a percentage and helps determine whether a match covers a substantial part of the query or only a short region. Query coverage is particularly important when comparing sequences of different lengths or identifying conserved domains within larger proteins.
- A simplified calculation is:
[
\text{Query coverage} =
\frac{\text{Number of query positions covered by the alignment}}
{\text{Total query sequence length}}
\times 100
]
- For example, if a query protein contains 500 amino acids and an alignment covers 400 query positions, its query coverage is 80%. If another alignment covers only 75 positions, its query coverage is 15%. The second alignment may still be biologically meaningful if it represents a conserved functional domain, but it does not demonstrate similarity across most of the query protein.
- Query coverage should not be confused with subject coverage, which measures the proportion of the database sequence covered by the alignment. A short query sequence may align across almost its entire length to a small region of a much larger subject sequence. In this situation, query coverage can be close to 100%, while subject coverage remains low. Some BLAST output formats report query coverage for individual alignments, while others provide values aggregated across multiple alignment segments or high-scoring segment pairs. Researchers should identify the exact definition used by the selected output field.
- The four measures provide complementary information. E-value describes the expected frequency of chance matches in the search space, bit score describes normalized alignment strength, percentage identity measures exact residue matches, and query coverage describes how much of the query is represented in the alignment. No single measure is sufficient for evaluating every sequence match. A strong interpretation considers all four together, along with alignment length, subject coverage, biological context, and the quality of the sequences being compared.
- For example, consider two hypothetical BLAST results for a protein query. The first database match has an E-value of (1 \times 10^{-50}), a bit score of 180, 65% identity, and 90% query coverage. The second has an E-value of (1 \times 10^{-8}), a bit score of 55, 95% identity, and 12% query coverage. The second alignment has higher identity, but it covers only a small part of the query. It may represent a short conserved domain or motif. The first alignment has lower identity but much stronger overall statistical support and substantially greater query coverage. Depending on the biological question, the first result may provide stronger evidence of similarity across the complete protein, while the second may identify a particularly conserved region.
- In BLAST, these statistics can be obtained using tabular output. The following command searches a protein query against a protein database and reports the subject identifier, percentage identity, alignment length, mismatch count, gap-opening count, query and subject coordinates, E-value, bit score, and query coverage:
blastp \
-query query_protein.fasta \
-db protein_database \
-out results.tsv \
-outfmt “6 sseqid pident length mismatch gapopen qstart qend sstart send evalue bitscore qcovs”
- The pident field reports percentage identity, evalue reports the E-value, bitscore reports the bit score, and qcovs reports query coverage per subject sequence. BLAST also provides other coverage-related fields, such as qcovhsp, which reports query coverage per high-scoring segment pair. These fields are not interchangeable, so the selected measure should match the intended analysis.
- A typical tabular result might look like this:
subject_001 65.00 450 140 3 20 469 15 464 1e-50 180 90
subject_002 95.00 60 3 0 10 69 80 139 1e-08 55 12
- The values are illustrative. The first result has 65% identity across a much larger aligned region and 90% query coverage. The second has 95% identity but covers only 12% of the query. This example demonstrates why percentage identity should not be interpreted without considering alignment length and coverage.
- These measures can also be processed programmatically. The following Python example reads a BLAST tabular file containing the fields shown above and displays the main statistics:
import pandas as pd
columns = [
“subject_id”,
“identity”,
“alignment_length”,
“mismatches”,
“gap_opens”,
“query_start”,
“query_end”,
“subject_start”,
“subject_end”,
“evalue”,
“bit_score”,
“query_coverage”
]
results = pd.read_csv(
“results.tsv”,
sep=”\t”,
names=columns
)
print(
results[
[
“subject_id”,
“identity”,
“evalue”,
“bit_score”,
“query_coverage”
]
]
)
# Sort by E-value, then by bit score
results = results.sort_values(
by=[“evalue”, “bit_score”],
ascending=[True, False]
)
print(“\nBest-supported matches:”)
print(
results[
[
“subject_id”,
“identity”,
“evalue”,
“bit_score”,
“query_coverage”
]
].head(10)
)
- When interpreting results, researchers should also consider sequence length, database composition, low-complexity regions, repetitive sequences, and the possibility of multiple domains. A short alignment can produce a high percentage identity by chance, particularly when it contains only a small number of residues. Conversely, a long alignment with moderate identity may provide strong evidence of evolutionary conservation. Low-complexity regions and biased sequence composition can also produce misleading matches, even when the E-value appears significant. Filtering methods such as SEG for proteins and DUST for nucleotide sequences can help reduce some of these effects, although filtering may also remove biologically meaningful regions.
- For protein searches, the substitution matrix and gap penalties influence alignment scores and therefore affect bit scores and E-values. BLOSUM62 is commonly used for general-purpose protein similarity searching, while other matrices may be more suitable for closely related or more distantly related sequences. Nucleotide searches use scoring systems based on match rewards, mismatch penalties, and gap costs. Consequently, comparisons across searches should account for differences in scoring parameters, search programs, database versions, and statistical settings.
- The significance threshold should also depend on the purpose of the analysis. A very low E-value is often required when identifying strong sequence matches, but more permissive thresholds may be useful during exploratory searches for distant homologues. In large-scale analyses involving thousands or millions of searches, researchers should consider multiple-testing effects and apply suitable statistical procedures where appropriate. An E-value threshold alone does not replace biological validation or correction for all forms of multiple testing.
- These statistics are widely used in genome annotation, protein function prediction, evolutionary analysis, orthologue identification, conserved-domain detection, pathogen identification, comparative genomics, and metagenomics. They help researchers distinguish strong full-length matches from short local similarities and prioritize candidates for further investigation. However, sequence similarity does not prove functional equivalence, and sequence homology refers to common evolutionary ancestry rather than a numerical percentage.
- For reproducible sequence analysis, researchers should record the query sequence, database name and version, search program and version, scoring matrix, gap penalties, filtering settings, E-value threshold, and output format. They should also retain the complete alignment when a result is important for downstream biological interpretation. Reporting E-value, bit score, percentage identity, query coverage, alignment length, and subject coverage together provides a more informative description of sequence similarity than relying on any one statistic.
- Overall, E-values, bit scores, identity, and query coverage are essential measures for interpreting sequence similarity searches. E-values assess the expected frequency of chance matches, bit scores quantify normalized alignment strength, identity measures exact residue matches, and query coverage indicates how much of the query sequence is represented. Their combined interpretation provides a stronger foundation for identifying meaningful sequence relationships and evaluating candidate biological functions.