One lab. 28,812 unique BSA/HSA QC injections from 2005–2026 (peaking at 4,178 in 2012), with 8,918 unique HeLa QC runs from 2017 onward (peaking at 1,785 in 2023). HeLa overtook BSA in 2020 as the dominant standard; a trickle of BSA continues. Spans 3D ion traps, LTQ-FT hybrids, Q-Exactive/Plus, Lumos, Exploris 480, timsTOF HT, and Astral. Counts use basename-strict cross-scan dedupe across Hive + DataArchive + Flinders (233,614 unique acquisitions total) and STAN's canonical HeLa pattern.
Scroll right to travel through two decades of instrumentation. Hover a card to see the era's best and worst runs.
Every circle is a BSA injection, every diamond is a HeLa run. Diamond size encodes QC sample load: small ≈ 50 ng (modern QC standard), medium ≈ 100 ng, large ≈ 500 ng. QE+ HeLa diamonds in 2017–2020 are mostly large/medium because the lab used 2–10× the modern 50 ng QC load on Q-Exactive Plus, which partly explains their higher ID counts. Click a point to see the run name. Y-axis is log-scale because today's HeLa runs produce 100× more identifications.
The same 607-amino-acid bovine serum albumin (P02769) molecule, characterized across instrument generations. Each bar shows exactly which peptides were identified — the story isn't just "more PSMs" but "deeper, more complete protein characterization."
File LTQ20060126bsa —
one of the earliest BSA injections in the archive with confident identifications.
Era-aware Comet v2026.01 search (2 Da precursor, 1.0005 Da fragment, rank-1
target-decoy 1% FDR) found 236 BSA PSMs with 35.7% sequence
coverage across 4,964 MS2 spectra. The Orbitrap era was still six years away.
File QEPlus2_07182018_41_hELA_500ng
(HeLa 500 ng) — 22,465 unique peptides /
41,614 Comet PSMs at rank-1 1% FDR. Best Q-Exactive
Plus HeLa run in our archive of 113 properly-FDR-controlled QE-HeLa
searches (median: 12,600 peptides / 19,097 PSMs). Apples-to-apples
with modern DIA-NN n_peptides from Lumos/Exploris/timsTOF.
Five injections in the archive are explicitly labelled
Michrome —
the Michrom Bioresources HPLC system the UC Davis Proteomics Core used
in the mid-2000s before switching to EasyLC / Dionex. The 0-PSM result
on these files isn't a search failure: the mzMLs
contain 370 MS1 scans and 0 MS2 scans. These were
MS1-only LC-MS profiling runs (intact-mass surveys, no
fragmentation), not DDA shotgun. No MS2 = no peptide identifications
possible regardless of search engine.
File ft20110216_7bsa
(LTQ-FT all-time peak) hit 613 BSA PSMs at 1% FDR with 39.9%
sequence coverage (Comet v2026.01, 20 ppm precursor + 1.0005 Da fragment,
rank-1 target-decoy FDR) — strong for a 2011 hybrid FT and competitive with
early Orbitrap results. The LTQ-FT's Fourier-transform MS1 gave it a decisive
edge over pure ion-trap instruments even at 2006-2011 scan speeds.
BSA was the primary QC standard from 2005 onward, peaking at 4,178 unique runs in 2012 (Q-Exactive era). HeLa appears in small numbers from 2005 but stays under 25/year through 2016. In 2017 HeLa jumps to ~270 runs, ramps through 2018–2019, then dominates from 2020 onward. HeLa peaks at 1,785 in 2023. BSA didn't fully disappear — a small trickle continues into 2026. See the transition chart below for the year-by-year shape (basename-strict dedupe across Hive + DataArchive + Flinders, 233,614 total acquisitions).
113 unique Q-Exactive Plus HeLa raws searched with Comet rank-1 1% FDR
against the community human FASTA: median 19,097 PSMs,
peak 41,614 PSMs (QEPlus2_07182018_41_hELA_500ng).
Sage cross-check on a representative 114-min top15 run: 30,176 PSMs (q≤0.01)
vs 22,920 PSMs Comet rank-1 1% FDR — Sage finds ~30% more, consistent
with its looser PSM-level FDR.
Load was NOT 50 ng like modern QC. Per-filename HeLa load: 500 ng (n=45), 100 ng (n=36), 250 ng (n=1), 50 ng (n=2), unmarked (n=29). The 41,614-PSM peak was 500 ng — 10× the modern QC load. Gradients ran 60–120 min vs the modern 14–35 min QC standard. Apples-to-apples comparison lives in the throughput violin below (peptides per minute of LC).
Year-by-year count of BSA/HSA vs HeLa QC raws across the lab's archives
(Hive historical_bsa + hela_qcs, DataArchive
Old-MassSpecData, Flinders mount, and to-hive staging).
Two trends visible: a steady BSA cadence from 2005 with peaks in 2012-2013,
a sharp tail from 2018, and the rise of HeLa QC from 2017 onward that
becomes the only standard by 2021.
Year extraction: filename YYYYMMDD/MMDDYYYY/MMDDYY/month-tag patterns first, file mtime as last resort. A small fraction of legacy files have ambiguous dates and fall into the file-copy bucket rather than the real acquisition year.
Year-by-year QC raw-file count per instrument family. Filtered to BSA /
HSA / HeLa raws using STAN's canonical patterns (bsa,
hsa, albumin, hela, hel50,
hel_, hel-, hel0, _hel),
deduped by basename across Hive + DataArchive + Flinders. 37,730 unique
QC acquisitions total. Each instrument is its own line so you can see
when each model started QC service and how long it stayed in active QC.
Y-axis: unique peptides identified per minute of LC gradient — the apples-to-apples scan-rate metric across DDA (Q-Exactive Plus, Comet rank-1 1% FDR) and DIA (Lumos / Exploris / timsTOF HT, DIA-NN n_peptides at 1% FDR). One violin per instrument; every dot is a real run. The bar inside each violin is the median; the box is the IQR.
Each bubble is one real run. X = gradient length (min), Y = unique peptides at 1% FDR, bubble size = ng HeLa load, color = instrument. Four dimensions in one chart. The fairest single-number summary is efficiency = peptides / (ng × min) — controls for both load and gradient. Median efficiency: Q-Exactive Plus 1.2, Lumos 9.4, Exploris 480 11.8, timsTOF HT 39.0 peptides/(ng·min). All-time record: 75.5 peptides/(ng·min) on timsTOF HT 100 SPD (41,542 peptides / 50 ng / 11 min).
Every dot is one real run across the lab's four active platforms (Q-Exactive Plus, Lumos, Exploris 480, timsTOF HT). Q-Exactive Plus spans 100 / 250 / 500 ng loads (n=82). Modern Lumos, Exploris 480, and timsTOF HT are all pinned at the community 50 ng standard. The chart makes it obvious the QE+ needed 2–10× more sample to reach lower peptide counts than today's instruments hit at 50 ng.
Real per-year counts of QC runs run in DDA vs DIA mode on each active instrument, pulled from the live STAN database. timsTOF HT (diaPASEF / ddaPASEF) and Orbitrap (DIA / DDA) modes are normalized into DIA and DDA. The flip from DDA-default to DIA-default happened in 2024–2025 across all three modern instruments.
Sample masses are not equal across eras: QE-HeLa 2017–2020 was typically 100–500 ng on 25 cm columns; modern Lumos/Exploris/timsTOF are 50 ng on shorter columns. Modern scan-rate gains are partly instrument speed (Astral, timsTOF HT, OT-IT parallel) and partly lower-load LC sharpness. QE-HeLa filtered to 18 of 113 runs where gradient length could be parsed from the filename.
A direct comparison of what the same core lab was measuring across instrument generations. Modern instruments use HeLa digest (more complex, 70,000+ proteins); the BSA numbers are comparably scaled to 50 ng / ~60 min gradients where available. Astral rows are real internal lab data from the UC Davis Proteomics Core's community speclib build — not a published reference.
| Instrument | Year | Sample | PSMs / Precursors | Coverage / Peptides | Relative depth | Source |
|---|
* Modern instrument medians from live STAN database (Lumos n=295, Exploris n=119, timsTOF HT n=315+).
* timsTOF HT 100 SPD precursor range from STAN community benchmark cohort.
‡ Astral numbers from UC Davis Proteomics Core internal speclib build
(comparison_astral_vs_lumos.tsv): 60 min HeLa 50 ng DIA ≈ 30k precursors,
120 min HeLa 50 ng DIA ≈ 44k precursors. Real lab data, same wet-lab pipeline as
every other row.
Both panels are real base-peak chromatograms extracted from
raw vendor files. Left: LTQ2007090702bsa_070912062529.RAW
(Sept 12, 2007 — 3D ion trap, BSA, ~49 min gradient). Right:
8jan25_HeL50-aftrPM-Dia_100spd_S4-A3 (Jan 8, 2025 — timsTOF
HT, HeLa 50 ng diaPASEF, 100 SPD). Note: not an apples-to-apples sample
comparison — modern instruments are run on HeLa digest, not BSA —
the point is peak density, FWHM, and elution complexity. A 2007 ion trap
resolves a dozen BSA peptide groups across 49 min; a 2026 timsTOF HT packs
~80 peptide elution events into ~11.5 min with sub-15-second FWHM and ion
mobility separation on top.
Real base-peak chromatogram from
090105HSA02.RAW — Human Serum Albumin (HSA, not BSA)
on a Thermo LCQ Deca XP 3D ion trap, September 1, 2005. This is the
oldest working raw in the museum (~20 years old) and a beautiful example
of what a 3D ion trap could deliver: peak ~3.63E8 (true Y axis in the
vendor view), MS1=1,505, MS2=216, with 45 HSA PSMs at 1% FDR and 41
passing the Yates/Washburn charge-state xcorr cutoffs — top
Comet xcorrs 6.09, 5.90, 5.01. The shape closely matches what Xcalibur
drew on screen 20 years ago: an early elution cluster at 10–12 min,
a dense midpeak at 14–18 min, and tail features past 22 min.