Protein annotation leaderboard
Compare two columns
LeaderboardTool cardsCompare two columnsHow the scores workRegion pile-upMethodsClaimsCoverage

Compare two columns

Any two columns of the leaderboard, drawn against each other: a score against what the search cost, or one score against another. Pick the column for each axis below; the ? beside each gives its definition.

Data from 15 Oct 2026, commit 4a4b1171 (source)

Change what is plotted

The benchmark split the proteins along one of the last four rows at a time, so picking a value in one of them sets the other three back to "any".

Used only when the horizontal axis is the estimated cost. Default prices are AWS on-demand rates for US East (N. Virginia), Linux, read 2026-10-15 from the EC2 on-demand price page: $0.044625 per CPU-hour (a c7i.large, $0.08925 per hour for 2 vCPU) and $1.006 per GPU-hour (a g5.xlarge with one NVIDIA A10G).
Figure 3. One marker per tool and setting, for the filters above. Hover or focus a marker to see its row; click it to open the tool's card.
phmmerMMseqs2jackhmmerMMseqs2, three roundsHH-suite (hhsearch)FoldseekProstT5 + Foldseekkmerseek, by alphabet:hp_thomas_dill2gbmr4wwmj5protein20hmmscan with Pfam no other tool is both cheaper and better

Colour is the kind of method; within a kind, marker shape is the tool, and for kmerseek the alphabet. Every cost here was measured on Sherlock; none was reported by a submitter. Core-hours, cost and wall-clock time are per 1,000 queries for searching one relative, the mean over the relatives each tool searched, measured on Stanford's Sherlock cluster. Peak RAM is the largest memory any one search task used. Wall-clock time is on the cores each tool was given (between 4 and 16; hover a marker for the number), so it can rank tools differently from core-hours. The database build is the mean over the builds the run's task records hold; a database reused from an earlier run left no record. Definitions: Methods.