Protein annotation leaderboard
Tool cards
LeaderboardTool cardsCompare two columnsHow the scores workRegion pile-upMethodsClaimsCoverage

kmerseek

Short words of k letters (k-mers) shared between two proteins.

Will it work for you?

  1. Is the closest same-family domain in the relative you will search at least 40% identical, for most of your proteins? If yes, Fmax here is 40–60% 0.005 (n = 563), 60–100% 0.297 (n = 1,289). If no, 20–30% 0.000 (n = 93), 30–40% 0.000 (n = 165). (Pfam truth set, raw calls, zebrafish (430 Mya) as the relative searched, setting gbmr4, k = 21; subsets with at least 30 true instances.)
  2. Are the features you care about longer than 60 residues? If yes, Fmax here is 61–120 aa 0.208 (n = 963), 121–250 aa 0.098 (n = 504), 251 aa or more 0.050 (n = 277). If no, 16–30 aa 0.281 (n = 271), 31–60 aa 0.232 (n = 420). (Pfam truth set, raw calls, zebrafish (430 Mya) as the relative searched, setting gbmr4, k = 21.)
  3. Can you spend about 0.0552 core-hours per 1,000 proteins for each relative you search? That is the measured mean per relative; see "Cost per 1,000 queries" below.

Best Fmax and cost

Under the Pfam truth set, raw calls, zebrafish (430 Mya) as the relative searched, its best setting is gbmr4, k = 21 (gbmr4_k21_lcFalse): Fmax 0.192 over 2,435 true instances. Searching costs about $0.00246 per 1,000 queries (0.0552 core-hours), for each relative searched.

Where it is strong and weak

Each line compares this setting with the best setting of any other tool on the same subset of the proteins, so the other tool's setting can change from line to line. Only subsets with at least 30 true instances are compared.

Where it is closest to, or furthest ahead of, the best other tool:

Where it is furthest behind the best other tool:

Swiss-Prot features, raw calls, zebrafish, setting wwmj5, k = 11: best feature type membrane-spanning, Fmax 0.362 (n = 1,124); worst coiled coil, Fmax 0.000 (n = 145).

Figure 2. Where kmerseek is ahead and behind: its Fmax on each subset of the proteins, against the best other tool on that subset (Pfam truth set, raw calls, zebrafish as the relative; its best setting, gbmr4, k = 21). Left out, under 30 true instances: 0–20%.
kmerseek best other tool on that subset, named at the right
% identity to the closest same-family domain20–30%best other tool: phmmer, default options, Fmax 0.359phmmerkmerseek: Fmax 0.000, n = 9330–40%best other tool: MMseqs2, three rounds, sensitivity 7, 3 rounds, Fmax 0.209MMseqs2, three roundskmerseek: Fmax 0.000, n = 16540–60%best other tool: phmmer, default options, Fmax 0.252phmmerkmerseek: Fmax 0.005, n = 56360–100%best other tool: phmmer, default options, Fmax 0.308phmmerkmerseek: Fmax 0.297, n = 1,289none in relativebest other tool: HH-suite (hhsearch), single-protein profiles, Fmax 0.032HH-suite (hhsearch)kmerseek: Fmax 0.000, n = 321length of the annotated region16–30 aabest other tool: HH-suite (hhsearch), single-protein profiles, Fmax 0.066HH-suite (hhsearch)kmerseek: Fmax 0.281, n = 27131–60 aabest other tool: phmmer, default options, Fmax 0.095phmmerkmerseek: Fmax 0.232, n = 42061–120 aabest other tool: phmmer, default options, Fmax 0.250phmmerkmerseek: Fmax 0.208, n = 963121–250 aabest other tool: jackhmmer, 3 rounds, Fmax 0.333jackhmmerkmerseek: Fmax 0.098, n = 504251 aa or morebest other tool: phmmer, default options, Fmax 0.477phmmerkmerseek: Fmax 0.050, n = 27700.10.20.30.40.5Fmax (higher is better)

kmerseek

claims/kmerseek.yaml, updated 15 Oct 2026
● 1 holds○ 3 not yet tested⊘ 2 disprovedlast disproved 13 Sep 2026

each row: the claim, the result that would disprove it, the verdict with its number, the evidence

● Holds (1)

○ Not yet tested (3)

⊘ Disproved (2)

claims.yaml schema ↗ badge and CI check ↗

What it compares

Short words of k letters (k-mers) shared between two proteins. First, each amino acid is replaced by a letter from a smaller alphabet, for example H for hydrophobic and P for polar. Two proteins can then match where their residues differ but their pattern of residue types is the same.

What it needs

Reference searched

An index of the relative's proteome (Quest for Orthologs reference proteomes, release 2020_04), built once per alphabet and k-mer length.

How to read its score

region_enrichment: the number of k-mers the region shares with the target divided by the number expected by chance (fold enrichment). Higher ranks first. A region was kept when either test passed: its own score was at least 1.3, or the whole query protein's p-value was at most 0.05.

Known failure modes

Cost per 1,000 queries

Setting gbmr4_k21_lcFalse, searching one relative (mean over the 15 relatives it searched): 0.0552 core-hours reserved per 1,000 queries (4 cores for the task's wall time), measured on Sherlock. Peak memory 5.2 GB. That is about $0.00246 at $0.044625 per CPU-hour. Building its databases took 0.147 core-hours over 6 databases. A database is built once and reused, so this is not in the per-query cost. This covers the 6 database builds the task records hold; the search used 15 relatives.

The per-query cost counts the search only. The run recorded cost per tool and setting, summed over all relatives, so the cost for one relative is the mean.

Licence

Settings in the table

Version, container, command

kmerseek search \
    --alphabet hp_thomas_dill2 \
    --ksize    24 \
    --query    UP000005640_9606.fasta \
    --target   zebrafish.hp_thomas_dill2.k24.lcfalse.kmerseek.rocksdb \
    --threshold         0.0 \
    --min-shared-kmers  2 \
    --max-query-pvalue  0.05 \
    --min-region-score  1.3

Shown for setting hp_thomas_dill2 k24. Other settings change --alphabet and --ksize only.

Tool id kmerseek. Numbers from seanome/2024-kmerseek-analysis commit 4a4b1171.

All tool cards · this card as Markdown · as JSON