We have developed a novel algorithm termed GoldenHaystack (GH) that was designed for enhanced peptide quantification of data-independent acquisition liquid-mass spectrometry (DIA-LC-MS) data files regardless of whether the amino acid sequences are subsequently assigned to the quantified peptide. The two central ideas behind GH are: (a) for sufficiently sized projects (e.g., ≥∼30 LC-MS files), pairs of peptides that coelute exactly in one subset of LC-MS files do not necessarily coelute exactly in a different subset of files, and (b) the ion intensity ratios between MS2 ions for any given peptide tend to stay the same across samples, but the ion intensity ratios of MS2 ions between different peptides tend to differ substantially across different samples. GH thus analyzes a project holistically: It uses multi-partite matching to match both MS2 (primarily) and MS1 (secondarily) ions across all samples, separates and regroups the MS ions into unique analyte quantifiable signatures (UAQS), reduces stochastic noise, and then quantifies those UAQS. In this paper, GH is compared to DIA-NN, a common algorithm used in DIA-MS proteomic analysis, and we demonstrate that GH (a) quantifies and identifies with better FDR accuracy known peptides found in FASTA search spaces (∼5-25% of analytes in DIA-MS data sets), (b) quantifies the remaining ∼75-95% of unassigned peptides that would be typically unquantified and unreported, and (c) runs ∼40-200× faster (or ∼1-10× faster than the LC-MS). Specifically, without a FASTA or spectral library, GH can deconvolute and accurately quantify chimeric LC-MS spectra. The use of a FASTA file occurs during an optional peptide identification step and is deployed only after the analytes in the MS files have already been quantified. We provide details of GH performance on several existing proteomics data sets, including plasma, cerebrospinal fluid, and cells.
Saxena et al. (Wed,) studied this question.