Skip to main content

Table 1 Performance comparison in terms of filling the matrix, traceback, conversion to CIGAR strings, total computation time, and GCUPS

From: Introducing difference recurrence relations for faster semi-global alignment of long sequences

Calc. time (sec.)

Fill

Trace

Conv

Total

GCUPS

Editdist

0.436

0.104

0.076

0.616

7.19

Non-diff

0.565

0.399

0.073

1.037

2.77

Diff-raw

0.516

0.316

0.073

0.905

3.03

Libgaba

0.377

0.097

0.028

0.502

4.15

Edlib

26.0

18.8

0.109

44.9

13.23

SeqAn ED a

77.2

   

7.97

BWA-MEM global b

354

0.381

 

355

0.12

BLAST X-drop c

250

   

0.18

Parasail c

886

   

0.69

  1. Results of comparison to existing implementations. Four adaptive banded DP implementations of ours (editdist, non-diff, diff-raw, and libgaba; top four rows) and four existing implementations (edlib, SeqAn ED, BWA-MEM global, BLAST X-drop, and Parasail; bottom five rows) were compared. See the main text for the details of the implementations. Columns: The average computation time per cell is shown in the Fill column. The time for traceback and CIGAR string conversion is shown in the Trace and Conv columns. The Total column presents the sum of Fill, Trace, and Conv. The GCUPS column shows the matrix fill-in performance in billion cell updates per second (GCUPS)
  2. aThe traceback and path-to-CIGAR conversion time for the SeqAn ED were not measured
  3. bThe path-to-CIGAR conversion field for the BWA-MEM global is blank because the implementation directly generates final CIGAR string
  4. cThe traceback and the path-to-CIGAR conversion fields are blank because the traceback routines were not implemented for the BLAST X-drop DP and the Farrar’s algorithm implementations that we used