What Is Sequence Homology and How Do You Infer It from BLAST Results?

Updated

Sequence homology is a conclusion about shared ancestry, not a measurement. Two sequences are homologous if they descend from a common ancestral sequence; you cannot assign a "percent homology." What BLAST gives you is sequence similarity — local regions where a query and a database sequence align well — plus statistics (E-value, bit score) that tell you how unlikely that similarity is to arise by chance. You then use that evidence, together with alignment coverage and biological context, to decide whether a homology claim is defensible. This applies to nucleotide and protein searches against public databases via NCBI BLAST.

Similarity vs. homology: keep the two separate

  • Similarity is observable and quantitative: percent identity, alignment length, gaps, bit score.
  • Homology is an inference about history: either the sequences share an ancestor or they do not. There is no such thing as "40% homologous."
  • High similarity is evidence for homology; it is not the same statement. Two short, low-complexity regions can be highly similar by chance, and two true homologs can diverge until similarity is faint.

BLAST's own description frames this correctly: it "finds regions of local similarity between sequences" and "can be used to infer functional and evolutionary relationships between sequences as well as help identify members of gene families." The inference step is yours.

What BLAST actually computes

BLAST compares your query against a sequence database and reports local alignments — the best-matching subregions, not a forced end-to-end comparison. For each hit you get:

Output field What it tells you How to use it
E-value Expected number of hits this good or better by chance in a database of that size Smaller is stronger; it depends on database size, so compare within a search, not across databases
Bit score Alignment quality, normalized so it is comparable across searches Higher is better; useful when E-values are all effectively zero
Percent identity Fraction of aligned positions that match Only meaningful alongside alignment length and coverage
Query cover / alignment length How much of your query participates in the alignment Low coverage weakens a homology claim even at high identity
Gaps Insertions/deletions in the alignment Many gaps suggest divergence or a poor alignment

The E-value is the field most often misread. It is not a probability that the match is homologous, and it is not a p-value in the usual sense — it is an expectation over the whole database search.

Reading a BLAST result to judge homology

Work through the evidence rather than stopping at the top hit:

  1. Check the E-value in context. A very small E-value (for example, far below 1e-5 for a typical protein search) is strong evidence the similarity is not random. Borderline values need corroboration.
  2. Check coverage. If the alignment covers only a small fraction of your query, a high percent identity over that fragment may reflect a shared domain rather than full-length homology.
  3. Look at the alignment itself. Are the matches spread across the length, or concentrated in a repetitive or low-complexity stretch? BLAST masks low-complexity regions by default for good reason.
  4. Compare multiple hits. If several unrelated database sequences all align to the same region of your query with similar scores, suspect a common domain or a compositional bias rather than orthology.
  5. Bring in outside evidence. Conserved domain searches, synteny, or experimental annotation strengthen or weaken the inference. BLAST alone rarely settles a homology question.

Common pitfalls

  • Equating high identity with homology. A 95% identity over 30 residues is weak evidence; 30% identity over 300 residues with good coverage may be strong.
  • Ignoring database size. The same alignment yields a worse E-value in a larger database. A hit that looks significant in a small custom database may not be in a comprehensive one.
  • Treating the top hit as the true homolog. The top hit is the best-scoring match in that database at that moment, which is not the same as the closest evolutionary relative.
  • Forgetting that "homolog" needs a qualifier. Once you accept homology, specify whether you mean orthologs (speciation) or paralogs (duplication), because the functional implications differ.
  • Over-reading percent identity across species. Rapidly evolving or saturated sequences can hide true homology; profile- or structure-based methods may recover relationships that pairwise BLAST misses.

What you can do once homology is supported

  • Function annotation: transfer a known function from a characterized homolog to an uncharacterized sequence, with the caveat that function does not always track ancestry.
  • Gene family assignment: group sequences into families using consistent thresholds and coverage rules, and document them.
  • Evolutionary analysis: build alignments and trees from sequences you have justified as homologous — the homology call is the prerequisite, not the output, of phylogenetic analysis.

The practical rule: report similarity as measured, and state homology as an inference with the evidence and thresholds you used.