Benchmark run 2026-07-22

LFQBench · Orbitrap Astral AWS c7i.8xlarge 7.23 h

261,903 Precursors

17,831 Protein groups

6 Raw files analysed

Quantification Accuracy

LFQBench mixes human, yeast and E. coli proteins into two samples at fixed, known ratios. A perfect result lands exactly on the target line.

Human

182,199 precursors

target 0.0

-0.0337

deviation from target -0.0337 spread (MAD) 0.1054

E. coli

19,320 precursors

target -2.0

-1.9172

deviation from target 0.0828 spread (MAD) 0.2037

Yeast

50,216 precursors

target 1.0

0.9445

deviation from target -0.0555 spread (MAD) 0.1401

Horizontal axis: log₂ ratio between the two samples. The shaded band shows the median absolute deviation — how tightly individual measurements cluster around the target.

Per-file Results

Each of the 6 raw files, analysed independently. Mass and retention-time errors indicate calibration quality.

File	Precursors	Proteins	MS1 error	MS2 error	RT error	FWHM (RT)
`A_R1`	240,598	19,278	0.4038	9.0448	15.2878	2.8135
`A_R2`	238,545	19,166	0.4211	7.5509	16.1881	2.6834
`A_R3`	241,116	19,277	0.416	9.2781	16.5531	2.8096
`B_R1`	238,505	19,241	0.5367	6.9808	16.1193	2.6418
`B_R2`	243,552	19,422	0.3868	7.6413	11.0295	2.6721
`B_R3`	243,011	19,383	0.4238	7.6488	10.2377	2.6743

How this was measured

LFQBench is a community-standard benchmark for label-free quantification. Because the mixing ratios are fixed when the samples are prepared, the correct answer is known independently of any software — which makes it possible to measure accuracy rather than merely report output volume.

Dataset

Orbitrap Astral — 6 raw files: conditions A and B, 3 replicates each.

Ground truth

Expected log₂ ratios — Human 0.0, E. coli -2.0, Yeast 1.0.

Environment

AWS c7i.8xlarge, single run. Random seed fixed so the analysis is reproducible.

Scope

Measured on the main branch of the SynapSpec engine. Runtime reflects one run on shared infrastructure and will vary with hardware.

Benchmark run