Skip to main content


Figure 1 | BMC Research Notes

Figure 1

From: BarraCUDA - a fast short read sequence aligner using graphics processing units

Figure 1

The search space traversal for inexact alignments. A. There are 4 types of alignments for each base, namely exact match, mismatch, insertion and deletion. A mismatch alignment is performed by substituting the reference base G with the three other possible bases (A, T and C). Insertions are detected by deleting the base from the query sequence, and deletions by inserting all 4 bases into the query sequence. B. A BFS strategy: BFS starts from the first base (1) and stores all hits in daughter nodes (1, 1)...(1, 3) in memory (with shaded squares), then it chooses (1, 1) and expands it into (1, 1, 1) ... (1, 1, 3). With all the nodes in memory, the agent then evaluates (1, 1, 1) which returns an alignment followed by (1, 1, 3), a suboptimal alignment, which is the next best hit in memory. After that the agent proceeds to (1, 2) which has the same number of differences as (1, 1, 3). BWA does not process nodes with more than z + 1 differences, i.e. (1, 3), (1, 1, 2) with the default option. C. A DFS strategy: DFS chooses the best hit (1, 1) from (1), and subsequently chooses (1, 1, 1) which returns an alignment. Then the agent goes back to (1, 1) to reach (1, 1, 3) as the next best hit to return the sub-optimal alignment. After that, the agent returns to (1, 1), then (1) to reach (1, 2) (not shown). Like BWA, BarraCUDA skips nodes with > z + 1 differences by default.

Back to article page