BLAST — Basic Local Alignment Search

Understanding E-Values: What the Numbers Really Mean

Every BLAST hit carries an E-value (Expect value) — the number of alignments with a score this good you would expect to find by pure chance given the size of the database searched. It is not a percentage of similarity; it is a statistical estimate of how surprising the match is. The lower the E-value, the less likely the match occurred randomly, and the more confident you can be that the alignment is biologically meaningful.

Quick Reference Thresholds

As a rule of thumb: for a protein search, an E-value below 1e-3 is a solid homology signal; for a nucleotide search, expect much lower E-values (often 0.0) for true matches, because nucleotide space is far larger and small differences stand out more starkly.

The Percent Identity Trap

A common mistake is judging a hit purely by percent identity — the fraction of aligned positions that match exactly. Percent identity only tells you how similar the aligned stretch is; it does not tell you whether that stretch is meaningful. A short, accidental 20-amino-acid stretch with 100% identity will produce a high identity number but a large (weak) E-value, because the match region is tiny relative to what chance would produce. Always read E-value and alignment length together with percent identity. A strong hit has both high identity and a very low E-value over a substantial alignment — never rely on percent identity alone.

Why the Same E-value Changes With the Database

Because E-value is a function of database size, the same query can score differently on different databases — and even on different runs of the same tool as the database grows. A hit that is striking against a small, focused database will look weaker (higher E-value) against the much larger NCBI nr or nt, where more random matches are expected. This is why the bit score exists: bit scores are normalized and independent of database size, so they are the fairer number when comparing results between searches or between tools. When your results differ across tools, compare bit scores and alignment coverage rather than raw E-values.

Nucleotide vs. Protein: Different Rules of Thumb

E-value cutoffs are not interchangeable between query types. BLASTN (nucleotide query against a nucleotide database) is far more sensitive to small differences — every position matters — so you can demand vanishingly small E-values and near-perfect identity for meaningful hits; this is the mode to use for primer specificity checking. BLASTP (protein vs protein) is more permissive because amino acids tolerate conservative substitutions, so biologically real matches often carry less extreme E-values, and homologs with obvious divergence can still appear well above the strictest cutoffs. When you switch between tools that run BLASTX or tBLASTN, the translation changes the statistical framing, so apply the protein-style thresholds to those outputs.

This is exactly where a pseudogene like BRCA1P1 can hijack a search: a processed pseudogene shares a long stretch near-identical to its parent gene, so it will turn up with a very good E-value before the real BRCA1 transcript appears. Knowing how to read coverage rather than the score is what separates the false friend from the genuine hit. The worked example in the next section shows how to tell them apart.

Three Real BLAST Searches, Read the Way We Do

BLAST is easy to run and easy to misread. These three realistic cases follow the same workflow every time: run the search, then decide which hit you trust and why, rather than trusting whatever comes back first.

Case 1 — A Pseudogene Beats the Real Gene (BRCA1 vs BRCA1P1)

You want to confirm a primer or a sequence belongs to human BRCA1. You BLAST it against nt. The top hit comes back as BRCA1P1 — a processed pseudogene — with a superb E-value, while the genuine BRCA1 transcript sits a couple of rows down. BRCA1P1 is a near-full-length retrocopy of BRCA1 cDNA, so over a long aligned window it is almost identical; the E-value loves it.

The fix is a framing decision, not a longer search. Restrict the organism to Homo sapiens and the database to RefSeq RNA, then read query coverage and alignment length together. A pseudogene hit aligns across the exons but drops out over the introns the real transcript covers, so its coverage and length trail the true gene. When two top hits look near-identical on E-value, the discriminating number is coverage, not the score.

Case 2 — Colony PCR: Only One Band Should Come Back

You screened ten colonies for an insert and now want to confirm the sequence at the insertion site. BLASTN against the plasmid backbone should give a single, unambiguous hit: your insert. If several comparable hits come back, or the top hit skips the junction, the usual culprit is a low-complexity primer or a repeat-related primer. Here the E-value is near-zero for every candidate, so it no longer separates them; what matters is that exactly one hit covers the full junction with no gaps. A second hit with meaningful identity is a reason to re-check the clone, not to re-run BLAST.

Case 3 — Cross-species Homology Without a Sequence Name

You cloned a conserved protein domain from a non-model organism and want to know what was targeted. BLAST collects every homolog that shares the domain, and the E-value drops as domain length grows, so many true orthologs cluster above noise. Read the list of species: a consistent run of hits across mammals, or strong conservation in one clade, tells you more than any single entry. If only a few distantly related proteins appear at low E-value, the query is likely a short conserved motif rather than a full-length gene — a sign to BLAST each half of the sequence separately.

When BLAST Gives Unexpected Results

Almost every "wrong" BLAST answer is one of a small number of recurring mistakes, all easy to diagnose. The table below maps the symptom to the fix.

When the E-value is irrelevant because every hit clears it, stop comparing scores and check coverage, alignment length, species, and annotation instead. Those separate a meaningful match from a confident false friend.

This page connects to the NCBI BLAST API for comprehensive sequence alignment. BLASTN, BLASTP, BLASTX, and tBLASTN searches run against NCBI's public databases (nt, nr, RefSeq RNA) with organism-specific filtering, and a local exact-match mode is available for quick direct matches without external API calls.

How BLAST Works: E-Values and Alignment Scores

BLAST (Basic Local Alignment Search Tool) breaks the query into short words — typically 11 nucleotides for blastn, 3 amino acids for blastp — finds exact matches in the database, then extends them into full alignments. The E-value counts how many alignments at that score you would expect by chance, so lower means more significant; bit scores are database-size-independent and are the fair way to compare runs. As a rule of thumb, below 0.001 is significant, above 1 is plausibly random — the E-value table above gives the finer-grained reading.

Which BLAST Program Should You Use?

Tips for Better BLAST Results

To get meaningful BLAST results: (1) always filter low-complexity regions to avoid spurious matches, (2) use organism-specific databases when searching for homologous genes in a known species, (3) adjust the word size parameter — smaller words increase sensitivity but slow the search, and (4) check alignment coverage, not just E-value. A low E-value with short alignment coverage may indicate a conserved domain match rather than a true ortholog. For primer design specificity checking, use BLASTN with the RefSeq RNA database and expect perfect or near-perfect matches only.