BLAST Seed-and-Extend Dot-Plot (2D)

Query vs. Database

Seeding & Extension

Word size k 6

Match score 1

Mismatch penalty 3

X-drop threshold 8

Show background dot-plot

Live Readouts

Seeds found

Best score

Best local alignment

—

X-drop stops on best diagonal

Single-letter dot-plot match Exact word (seed), length k Best local alignment cell Score trace — X-drop reset point

How it works

BLAST never runs full dynamic-programming alignment over the whole query × database matrix — that would be too slow at genome scale. Instead it uses a three-stage heuristic:

1. SEED    index all length-k substrings ("words") of the query;
           scan the database for exact k-length matches → anchor points
2. EXTEND  from each seed, walk outward along its diagonal, adding
           match score (+m) or mismatch penalty (−p) at each step
3. X-DROP  stop extending once the running score falls more than X
           below the best score seen so far on that diagonal:
             stop when  bestSoFar − current(t) > X
           then trim the alignment back to the point of bestSoFar

The top panel is the classic 2D BLAST dot-plot: drag to pan, scroll/pinch to zoom. Exact k-length seed words light up on their diagonal. The bottom panel plots the running extension score along the single best-scoring diagonal as a line, with a shaded band showing exactly how far the score is allowed to dip (the X-drop threshold) before extension resets — this makes the stopping rule itself visible, not just its end result.

Word size k — shorter words seed more often (more sensitive); longer words seed rarely but only on strong homology.
Match score / mismatch penalty — the step size added or subtracted from the running score at each position of the extension.
X-drop threshold — how far the score may fall below its running best before extension gives up and resets; larger X tolerates longer stretches of mismatches inside one alignment.

The database sequence is built by taking a chunk of the query, mutating roughly 15% of its bases, and burying it inside random flanking DNA — exactly the kind of divergent homolog BLAST is designed to find.

Dot-plot — database → (x) | query ↓ (y)

zoom 1.0×

Extension score along best diagonal