BLAST and Database Searching

Core Idea

BLAST (Basic Local Alignment Search Tool) rapidly searches sequence databases by finding short exact matches (seeds), extending them into high-scoring segment pairs, and evaluating statistical significance using E-values. Unlike exact Smith-Waterman, BLAST uses heuristics to achieve speed suitable for searching millions of sequences. The E-value quantifies how many alignments of equal or better score would be expected by chance in a database of that size, making it the primary filter for distinguishing genuine homology from random similarity.

How It's Best Learned

Run a BLAST search at NCBI with a known protein sequence, then interpret the results: sort by E-value, examine the top hits, and check whether the aligned regions correspond to known domains. Repeat with a randomized version of the same sequence to see what background noise looks like.

Common Misconceptions

• A low E-value does not mean the sequences have the same function — it means they share significant sequence similarity, which suggests common ancestry but not necessarily conserved function.
• BLAST is not guaranteed to find the mathematically optimal alignment; it is a heuristic that can miss weak but real homologies.

Explainer

Searching a sequence against a database to find relatives is the bread-and-butter operation of bioinformatics. You have a gene or protein sequence and want to know: what is this? What organisms have something similar? What is its likely function? The Smith-Waterman algorithm gives the exact best local alignment, but running it against millions of database sequences would take days. BLAST, developed by Altschul et al. in 1990, solves this by trading guaranteed optimality for enormous speed gains through a clever heuristic strategy.

BLAST works in three stages. First, it breaks the query into short "words" (typically 3 amino acids for protein, 11 nucleotides for DNA) and identifies all database sequences containing exact or near-exact matches to those words. This seeding step is extremely fast because it uses precomputed lookup tables. Second, it extends each seed in both directions using ungapped alignment, stopping when the score drops below a threshold. Third, it takes the highest-scoring extensions and performs gapped alignment in a narrow band around them. This three-stage filter eliminates the vast majority of database sequences before any expensive computation happens.

The statistical framework behind BLAST results is what makes them interpretable. The key metric is the E-value (expect value), derived from Karlin-Altschul statistics. For any alignment score S, the E-value tells you how many alignments scoring at least S you would expect purely by chance in a database of that size with sequences of those compositions. An E-value of 0.001 means you would expect such a score by chance roughly once in every 1,000 database searches. Crucially, E-values depend on database size — the same alignment score produces a higher E-value in a larger database because there are more random comparisons being made, just as running more statistical tests increases the chance of a spurious result.

In practice, E-values below about 1e-5 are generally considered strong evidence of homology, while values between 1e-5 and 0.01 merit careful inspection. But BLAST results require biological judgment beyond the E-value. Two sequences may be clearly homologous (share common ancestry) yet have diverged in function. Conversely, BLAST may miss genuine homologs if the sequences have diverged so far that the seed-finding heuristic fails — a limitation that more sensitive methods like PSI-BLAST and HMM-based searches (HMMER) address by building position-specific profiles from multiple related sequences.