Contents

Evaluation conditions

Last updated: 21 July 2026

Results reflect the tool versions and specifications as of 21 July 2026. Tool updates may change the outcome.

We compared how many approximate match candidates GGGenome, BLAST and Bowtie detect for a short query sequence.

ED = mis + ins + del

ED is the edit distance: the total of mismatches, insertions and deletions. Here, ins denotes an insertion in the query relative to the target genome (a gap on the genome side), and del a deletion in the query relative to the target genome (a gap on the query side).

How detection is defined here

In this comparison, a “detection” is a sequence candidate that satisfies the search conditions configured for each tool and appears in that tool’s output for the specified 20-mer query.

Search settings for each tool

GGGenome — package edition v1.2.1

GGGenome performs exhaustive approximate search based on edit distance for short query sequences. It is available as a web edition from DBCLS and as a package edition from Retrieva. This evaluation used package edition v1.2.1 with the command below.

wget "http://localhost:xxxx/hg38.p14_d3g2302/3/GGGAGGCCCACGTATGGCGC"

BLAST — 2.14.1

BLAST is a heuristic local alignment method that extends alignments from word hits, available as a CLI download from NCBI.

blastn \
  -task megablast \
  -query VIM_20.fa \
  -db hg38.p14 \
  -out VIM_20_hg38.p14.csv \
  -word_size 4 \
  -evalue 10000 \
  -outfmt '10 std qseq sseq'

Bowtie — v1.3.1

Bowtie 1 is a short-read aligner based on a BWT/FM-index, distributed as a CLI by Johns Hopkins University.

bowtie \
  -f VIM_20.fa \
  -x hg38_index \
  -v 3 \
  -a

These settings output every ungapped alignment with up to 3 mismatches across the whole sequence. Because Bowtie 1 does not support gapped alignment, edit patterns containing insertions or deletions are out of scope under these conditions.

Normalizing the results

For each output alignment, genome sequence corresponding to the unaligned portion of the query was retrieved and an alignment covering the full 20-base query was reconstructed. Candidates whose total of mismatches, insertions and deletions was 3 or less were then extracted, and duplicates from the same chromosome, strand and genomic region were merged. Genomic coordinates shown here normalize each tool’s output to 1-based inclusive form.

Normalizing the BLAST output

The BLAST search under the settings above output 4,935 alignments. Because BLAST is a local alignment method, that output includes partial alignments that do not cover the full query, alignments beyond an edit distance of 3, and multiple alignments to the same genomic region. After normalizing to the common criteria, 255 were accepted as valid candidates. All 255 correspond to candidates that GGGenome also detected.

Example: chr1 11659146-11659162

BLAST output (1 mismatch and 1 insertion over 18 bases, evalue=3495)

After extension to the full 20-base query (3 mismatches and 1 insertion over 20 bases)

The edit distance here is 4, so this candidate was not accepted.

Results

Summary

Detected counts (under the search conditions used here)

Candidates with ED 1–3, excluding the exact match (ED 0).

Detailed results by edit pattern

Table 1. Search results for GGGenome, BLAST and Bowtie

N/A indicates the pattern is out of scope because Bowtie 1 does not support gapped alignment. “—” marks cells with no figure in the source material.

Detected counts for the main indel conditions (GGGenome / BLAST)

Why do GGGenome, BLAST and Bowtie return different results?

GGGenome, BLAST and Bowtie differ in their intended use, search algorithms and search conditions, so the same 20-mer query does not necessarily return the same number of sequences. Under the conditions used here, the following differences accounted for the gap in detected counts.

The difference in detected counts reflects which edit operations each tool supports and the search parameters set here, not the merits of the algorithms themselves.

Discussion

This difference appears to stem from the design intent of each search algorithm.

Conclusion

This evaluation used the 20 bases at the 5′ end of VIM transcript NM_003380.5 as the query against D3G Human genome GRCh38.p14, comparing the number of approximate match candidates within an edit distance of 3. For this query and these conditions, every method detected one exact match. Excluding the exact match, the candidate counts were 554 for GGGenome, 255 for BLAST and 81 for Bowtie, with GGGenome returning the most.

Under mismatch-only conditions the three tools agreed. Where indels were involved, BLAST returned fewer valid candidates than GGGenome. Because Bowtie 1 does not support gapped alignment, conditions involving indels are out of scope for it.

Frequently asked questions

Q. Why do GGGenome, BLAST and Bowtie return different numbers of results?

A. Because each tool uses a different search algorithm and different search conditions.

Even with the same input sequence, the number of results changes with how mismatches and gaps are handled, the alignment method, and the output conditions. Under the conditions used here, GGGenome enumerated candidates including insertions and deletions on the basis of edit distance, BLAST performed local alignment extended from word hits, and Bowtie 1 does not support gapped alignment, so conditions involving indels were out of scope.

Q. Does this result mean GGGenome is more accurate than BLAST or Bowtie?

A. No.

What is compared here is the number of 20-mer approximate matches detected under the conditions set out on this page. It is not a direct measure of general search accuracy, sensitivity or processing speed.

Q. Which tool is suitable for searching off-target candidates?

A. It depends on the search space, the sequence length, the purpose of the search, and the conditions you need.

For the conditions in this evaluation — exhaustively enumerating candidates within an edit distance of 3, including insertions and deletions, for short sequences of around 20 bases — GGGenome is applicable. If the number of mismatches allowed, the need for gaps, or the database searched changes, the appropriate tool may change as well. Please contact us to discuss settings for your particular use.

Q. Do the results change if I change the sequence length or the conditions?

A. Yes.

Results vary with the sequence length, the number of mismatches allowed, gaps and the database searched. The figures on this page are the results for the single query and set of conditions described.

References

1. GGGenome, Help / Database Description, DBCLS.
   GGGenome Help

2. Jan Fassler, Ph.D. and Peter Cooper, Ph.D., BLAST Glossary, July 14, 2011.
   BLAST Glossary

3. Ben Langmead, Cole Trapnell, Mihai Pop, Steven L Salzberg, Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 2009 Mar 4;10(3):R25.
   PMC2690996

4. Bowtie Manual, Bowtie 1 documentation.
   Bowtie Manual