How to Interpret BLAST Results: E-Value, Bit Score, Identity, and Coverage

Quick answer: No single number in a BLAST output decides whether a hit is biologically meaningful. The E-value tells you how statistically surprising a match is, the bit score tells you how good the alignment is on a common scale, percent identity tells you how similar the residues are, and query coverage tells you how much of your sequence the hit explains. A truthful BLAST interpretation reads these four together, plus the biology of your question.

Complements the two earlier BLAST guides on this blog: What Is BLAST in Bioinformatics and Choosing the BLAST Program. Here we focus on the result table itself.

Where your attention should start

Run a BLAST search and the results table greets you with columns of numbers: Query Cover, E value, Percent Identity, Score, Accession, and description text. Beginners often scan straight for the top row of percent identity. That instinct is half right: percent identity is useful, but it is not the number BLAST ranks by, so the top row of the table is not automatically your best answer.

BLAST orders hits by best score first, using the bit score. A long weakly-similar alignment can therefore out-rank a shorter, nearly perfect one, and a low-E-value hit for your real gene can sit below a flashy repeat match. So the skill you need is knowing which number answers which question:

• Is this match real, or could I see it by chance? — E-value
• How good is the alignment across the matched region? — bit score
• Are the residues matching at each position? — percent identity
• Does the hit cover the part of my sequence I am asking about? — query coverage

The E-value: how often you would expect this by chance

The E-value is the expected number of chance alignments with a score at least this good in a database of this size. Two properties follow directly from the definition, and they explain most confusion:

• It depends on database size. The same alignment gets a bigger E-value in NCBI nt (a gigantic database) than in your small local database, purely because more opportunities for chance matches exist.
• Smaller is better, and zero-ish is meaningful. An E-value of 1e-40 means roughly zero chance alignments that good should exist in a database this size. An E-value of 10 is noise territory — expect about ten chance hits at least as good.

A common rule of thumb for beginners: below 1e-5 is often considered statistically meaningful, but this threshold exists because someone tested it in a specific context. It is not biology. Two proteins from different families can share short motifs that trap you at loose E-values, and two sequences from recently duplicated genes can drift apart at genuine E-values. The E-value is a statistics layer; biology enters next.

The bit score: one scale across searches

The bit score rescales the raw alignment score (match, mismatch, gap penalties) into log-odds units, roughly meaning twice as many bits = four times as unlikely by chance. Practically, three facts make it useful:

• Higher is better, always, and it is comparable across database sizes better than the E-value is.
• It drives the ordering you see in the results page.
• The relationship bit score ≈ 1.4 × raw score grounds intuition, but the exact constant depends on the scoring matrix, so treat the bit score itself as the number to report.

When you write up BLAST results, give the bit score and the E-value together: they describe the same alignment quality from two directions (a purely statistical one and a normalized strength one).

Percent identity: how similar the matched residues are

Percent identity is the fraction of aligned residue positions that match exactly, usually reported over the aligned region. It answers how closely related the aligned slice is — but only for that slice, which is why it pairs with coverage:

• 100 percent identity over a short stretch can be a low-information motif.
• 85 percent identity across a full-length gene is usually a strong ortholog signal.
• Identity below roughly 25 percent between proteins is generally indistinguishable from random unless the alignment is very long.

A mismatch column and indel counts also matter, because transitions in a conserved domain can change function even when identity stays high. This is why a confident annotation often goes past the table into looking at the alignment itself.

Query coverage: how much of your sequence the hit explains

Query coverage is the fraction of your input sequence that the alignment spans. It matters because a hit can be born statistical and still be biologically wrong-scale:

• A short high-identity alignment covering 8 percent of your query might be a repetitive element, not your gene.
• A longer moderate-identity alignment covering 95 percent of the query is often your candidate.

Coverage pairs with a second metric very similar to it: the alignment length relative to the query length. When a hit’s coverage is low but E-value is strong, ask whether your query contains a region that genuinely matches something common in the database.

Reading them together: a worked example

None of the four rows above is automatically good or bad until you state the question. For primer design you may want row type three. For annotating a transcriptome, type two is the workhorse. That is what interpreting means: metric combined with intent.

Practical thresholds without pretending they are laws

Rules that work better than any single cutoff:

• Read E-value and bit score as a pair; trust an alarm only when both are weak.
• Harbor suspicion when identity is high but coverage is tiny — check what the matched region actually is.
• For reporting, keep the accession number, percent identity, coverage, and E-value of the hit you rely on. A claim without those four is not auditable.
• When several database hits of the same species tie near the top, the deeper question is usually which biological form (isoform, strain, or assembly artifact) best matches — compare descriptions and relationships rather than the single number.

BLAST itself also warns that short queries need adjusted word sizes and statistics. If your sequence is short, generic defaults distort everything above, so this applies most to longer sequences analyzed with standard parameters.

Common mistakes that cost hours

• Treating the top-ranked hit as the truth without reading its description. Ranking can put a fragmented assembly or an over-annotated entry at the row you glance at first.
• Comparing E-values across searches with different database sizes. A 1e-50 in nt is not the same claim as a 1e-50 against a 1,000-sequence trial set.
• Chasing the lowest E-value blindly with high-identity short stretches. Match length interacts with significance; confirm coverage.
• Ignoring the alignment view. Columns of gaps and mismatch clusters reveal where the biology actually disagrees, which a single percent does not.

# What to record when you summarize a BLAST hit (pseudo-notebook cell)
record = {
  'accession': 'NP_000508.2',   # so anyone can reproduce
  'identity': 93.0,             # percent over aligned region
  'coverage': 96.0,             # percent of MY query
  'evalue': 2e-79,              # database-dependent chance count
  'bit_score': 412,             # normalized alignment strength
}

Key takeaways

• The E-value counts expected chance matches; it shrinks with better alignments but grows with database size — never compare E-values across different databases.
• The bit score ranks the results and normalizes alignment strength on one scale.
• Percent identity describes the matched positions; query coverage describes how much of your sequence the alignment explains.
• Evaluate identity with coverage, and evaluate both with significance: a hit is meaningful when metrics agree with each other and with biology.
• Report accession plus the four numbers of the hit you rely on, so readers can reproduce the claim.

Challenge

Run any sequence through NCBI BLAST, deliberately pick the fourth-ranked hit, and state for it: bit score, E-value, identity, coverage, and one sentence about which biological explanation (ortholog, shared domain, repeat, or merely weak similarity) fits those four numbers. If you cannot, pick the metrics you did not consider and re-read the hit before deciding.

Want help interpreting results in a course or research project? Ampersand Academy offers one-to-one mentoring in bioinformatics and data analysis — part of Mahadhi, a development and digital marketing company.

—

FAQ

Coverage, identity, and significance answer different questions; no single metric decides. In practice the combination of a strong bit score, a very low E-value, and coverage high enough to match the region of interest is what most people mean by a good hit.

• Which metric matters most in a BLAST result?

The often-quoted cutoff near 0.05 is a convention for treatments, not a biological threshold. Interpretation depends on database size, query length, and whether the aligned region has biological sense.

• Is an E-value below 0.05 always meaningful?

The E-value depends on database size and search parameters. A hit against a small trial set will show different values from the giant NCBI nt database even with the identical alignment.

• Why do E-values change between two searches of the same hit?

Usually not, or only partially. A perfect but tiny alignment often marks a repeat, a primer region, or a shared domain; check the matched region before concluding.

• A hit has 100 percent identity but low query coverage. Is it my gene?

Both describe the surprise level, but the bit score is normalized and comparable across database sizes better than the E-value, and it is what determines the ranking order.

• What does the bit score add beyond the E-value?

Ranking is by score, not by biology. The top hit can be a fragmented database entry or a different isoform; read descriptions and relationships, not just the first row.

• Should I always pick the top-ranked hit?

Identity alone hides where differences cluster. A conserved protein can hide one mismatch inside an active-site motif, so inspect the alignment view before making a functional claim.

• How do mismatches and gaps affect interpretation?