Standardized proteomic Throughput ANalyzer
Compare your QC HeLa against reference ranges from labs running the same frozen search.
Enter one QC run and how it was searched. A count from the community search (DIA-NN 2.3.x against the frozen community library at 1% FDR; for DDA, Sage 0.14.x against the frozen FASTA) is placed in its cohort: the same QC standard, instrument model, acquisition mode, gradient and amount as the reference ranges below. A count from any other search is not converted. You get the reason, and what would make it comparable.
Longitudinal performance ranges from the runs submitted so far, grouped by instrument model. Each card is one cohort: the same model, acquisition mode, LC and gradient, and load. Evosep runs are named by their Evosep method; nanoLC runs by the gradient their samples per day imply (1440 ÷ (1.25 × SPD)) and then the run length recorded with them. The large number is the cohort's median precursors (DIA) or PSMs (DDA) at 1% FDR, and the bar shows the middle half of its runs inside the 10th–90th percentile, on one scale per instrument. Every card says how many runs and labs it holds; a card built from one lab's runs is marked single-lab reference. Below 10 runs a card lists its values instead of a range. A cohort is not ranked, and is folded under its instrument with the reason, when it has fewer than 5 runs, records no SPD, records no LC at an SPD that is also an Evosep method, or records Evosep at an SPD that is not an Evosep method. Proteins are context only. The filter bar above chooses what is shown.
Showing
Any lab running STAN can add its QC HeLa runs. Your lab appears under a pseudonym you claim, and your runs join the ranges above.
display_name in community.yml (~/.stan/community.yml on macOS and Linux, %USERPROFILE%\STAN\community.yml on Windows), or run stan setup; stan init leaves the name empty, and stan community-claim refuses to run without one. Then run stan community-claim (STAN 1.2 or later) and enter the 6-digit code it emails you (from noreply@stan-proteomics.org; check the spam folder). Do this first: an unclaimed name can be taken by anyone.community_submit: true in the same community.yml, then run stan submit-all to send the runs you already have (or press Sync on the dashboard's Community tab, which does both). Schedule stan submit-all to keep new runs flowing.Your runs appear after the nightly rebuild at 04:00 UTC. STAN searches them with the pinned engines and frozen parameters described under Methods. Comparisons with the runs already here are close but not yet exact: nothing verifies which library a run was searched against, and UC Davis's timsTOF HT and Exploris 480 cohorts used subset libraries until a planned re-search (see Methods).
From /api/leaderboard, plus the two TIC fields from /api/tic-overlay.
Runs on this page are searched with pinned engine versions and frozen parameters, against one frozen FASTA and, for DIA, a per-vendor HeLa library. This is what that means in practice, and where it does not hold yet.
Precursor.Id (modified sequence + charge) at run-level Q.Value ≤ 0.01Stripped.Sequence among those precursorsProtein.Group with PG.Q.Value ≤ 0.01 among those precursorsPrimary: precursors (DIA) and PSMs (DDA), the purest instrument signal. Secondary: peptides. Proteins: context only; not used for leaderboards; 20% of IPS. Health: the ID-free charts below the Explorer (MS1 mass accuracy, MS1 signal, dynamic range and points across peak).
One frozen FASTA and empirical HeLa libraries, one per vendor (timsTOF ~54k, Orbitrap ~170k precursors), built from real HeLa DIA runs. Download them from the dataset.
The checksums on a row do not prove what was searched. For every DIA-NN 2.3.x run, STAN stamps the frozen checksums above onto the submission; it does not hash the FASTA and library the search actually used, and a row is marked assets-verified whenever a checksum is present. Nothing yet checks which library a lab searched: an install that has built its own instrument_library.parquet is searched against that automatically.
Not yet true for UC Davis's timsTOF HT and Exploris 480 cohorts. Those runs were searched against subsets of these libraries built from the lab's own runs (about 51,000 and 53,000 precursors), which no outside lab has, although their rows carry the full library's checksum. On the same timsTOF raws the full library gives about 1.034× the subset's count. A re-search of those cohorts against the full frozen libraries is planned (decision 11). The Fusion Lumos cohorts already use the full Orbitrap library.
Counts compare within a vendor, not across: timsTOF and Orbitrap runs search different libraries, and above 90% library coverage the library rather than the instrument caps the count. Coverage divides by the full library's size, so for runs searched against a subset it reads slightly low.
IPS places a run's identification depth against a fixed calibration set for its instrument and SPD: 60 is the median calibration run and 90 its 90th percentile. It weights 50% precursors (PSMs for DDA), 30% peptides and 20% proteins. The calibration set is 359 UC Davis HeLa QC runs from April 2026.
IPS is not shown on this page yet. The stored community scores come from a reference lookup that scored Exploris and timsTOF runs against a pooled all-instrument reference instead of their own, so Exploris scores read too low and timsTOF scores too high. Even keyed correctly, typical runs on both sit near 50–54 rather than 60, so the April 2026 references also need recalibrating. Scores return here once they are recomputed and recalibrated.
All STAN benchmarking uses the Pierce HeLa Protein Digest Standard (Thermo Scientific, cat# 88328 / 88329): a tryptic digest of the HeLa S3 cell line with >15,000 proteins and <10% missed cleavages. One commercial standard means differences reflect the instrument and LC, not sample preparation.
The number of MS2 scans sampling each precursor's elution profile limits how accurately it can be quantified. STAN's guidelines: 12+ points reliable quantitation; 6–12 points acceptable, depending on peak shape; <6 points quantitation error grows quickly. At high SPD with short columns, cycle time can exceed peak width. These are guidelines: actual error depends on peak symmetry and integration method.
Background: Matthews & Hayes, Anal. Chem. 1976, 48, 1375–1382, on how few sampling points bias GC-MS ratio measurements (doi:10.1021/ac50003a028).
Each point is one QC run. Every chart here follows the filter bar at the top of the page and says in its badge what it shows. A chart that deliberately leaves a filter out says so: Depth by Amount Loaded shows every amount, Depth by Throughput and Throughput vs. Quantitation Quality every gradient, and Column Comparison every column. Under "Both", DIA and DDA keep their own table, violin or facet, with one exception: Depth by Amount Loaded puts precursors and PSMs on one axis, and its badge warns about it. The TIC overlay follows the QC standard and DIA / DDA (never both at once) and keeps its own SPD and LC menus. Counts compare fairly within a vendor: timsTOF and Orbitrap runs search different libraries (see Methods).
Ranked within a vendor only: timsTOF runs search a ~54,000-precursor library and Orbitrap runs a ~170,000-precursor library, so a timsTOF row above an Orbitrap row is not a verdict on the instruments.
Compare violins of one vendor only: timsTOF and Orbitrap runs search different libraries (~54,000 vs ~170,000 precursors).
Compare facets of one vendor only: timsTOF and Orbitrap runs search different libraries (~54,000 vs ~170,000 precursors), so a timsTOF box above an Orbitrap box is not a verdict on the instruments.
These are the metrics the 2024 proteomics QC literature (NIST MSQC, QCloud2, CPTAC, PTXQC) considers the real LC-health signals: they catch failures before identifications collapse, and they don't depend on how many peptides you identified. One colour per instrument model; each dot is a run and the line is that model's monthly median. Click a legend entry to hide a model. The charts follow the filter bar; mass accuracy, MS1 signal and dynamic range come from DIA-NN, so DDA runs do not carry them.
Your lab against the community, one cohort at a time: "Is my instrument drifting, and where do I sit against other labs running the same thing?" The gold band is the lab's own baseline: the median ± 3 robust SD (1.4826 × MAD) of its first 30 runs in the cohort. It is drawn from 20 runs, provisional until there are 30, and fixed from then on, so a problem shows up even when no other lab shares the cohort, and a bad stretch is flagged instead of widening the band. The white line is the median of the last 15 runs, so slow drift is visible too. Other labs in the same cohort, never the lab itself, appear as grey 10th–90th and 25th–75th percentile bands once they have 5 runs in the window. Labs and cohorts are listed from the filter bar's view: labs with 5 or more runs in a ranked cohort.
Polyethylene glycol (PEG) from tips, solvents and plastics shows up in MS1 as a ladder of ions 44.026 Da apart, and it can cost identifications. STAN measures it in QC runs. Evosep labs that opt in share their per-run PEG here, so you can see whether your level is normal and whether a fix worked. PEG needs no database search, so this board is separate from the benchmark above and the filters at the top of the page do not apply to it.
Compare LC systems within one instrument family only. PEG share depends on the detector (STAN counts MS1 peaks above an absolute 104 intensity floor), so comparisons across instrument families, such as an Evosep timsTOF against a nanoLC Orbitrap, are not like-for-like.
Any lab running STAN 1.2 or later can join. PEG is read straight from raw MS1, so it needs no community search.
stan community-claim to prove it with an emailed code. Do this first: an unclaimed name can be claimed by anyone, who then takes over its place on the board. Unclaimed names show as unverified.peg_share: true to ~/.stan/community.yml. Sharing stays off until you do.stan peg-sync. It sends every QC run that has a PEG measurement, and the relay keeps only what changed, so it is safe to run again at any time.Lab names are pseudonyms. Shared per QC run: date, instrument model, LC system, Evosep method (SPD), acquisition mode, sample type and amount, and the PEG share, score, ion count and class. File and sample names, raw data, spectra and serial numbers never leave your lab.
Found PEG? Isolate the source in one night with STAN's Evosep PEG diagnostic.
PEG share of MS1. STAN reads 80 MS1 scans spread across the gradient and matches peaks within 5 ppm to the PEG ladder: PEG1–20 as [M+H]+, [M+NH4]+ and [M+Na]+, spaced 44.026 Da (C2H4O). The share is the matched intensity over all MS1 peaks above 104 counts. Unlike the 0–100 PEG score, it keeps rising with contamination, so it still separates labs that all score 100.
Cohorts. Labs are ranked only against the same instrument family and Evosep method, because detector response and gradient length change the number. Only Evosep runs are ranked.
Rank. Median over the window's QC runs, lowest first, with at least 5 runs; ties go to more clean runs, then more runs. Clean means a PEG score below 20. Runs with no PEG measurement are left out; they never count as clean.
Badges. Cleanest is rank 1 once two or more labs are ranked. Most improved is the biggest fall in median against the previous window, if it fell by 15% or more and by at least 0.5 percentage points, also once two or more labs are ranked. Change stays blank when the previous window had fewer than 5 runs or a median below 0.1%, where a percent change is noise.
All submissions are anonymous by default. Click column headers to sort. Use the export button to download as CSV. The table follows the filter bar; the DIA / DDA / All tab here is the bar's mode, so it sets the whole page. Each run's percentile is within its cohort, named in the Cohort column; a cohort that is not ranked shows a dash, with the reason. An amount marked assumed is the lab's default, not recorded for the run. Runs held back from the ranges and rankings (an amount unconfirmed or implausible) are listed below the table.