Proteomics MS: DDA vs DIA, Sample Prep and QC

Q: Which file formats should I archive?

Convert vendor raw files to mzML for the spectra, keep mzIdentML or mzTab for results, record SDRF-Proteomics for the sample-to-run mapping, and deposit to a ProteomeXchange member repository for a PXD accession. Archive the exact FASTA and search parameters alongside them; without those, the results are not reproducible even from perfectly preserved raw data.

A discovery proteomics experiment is decided long before anyone looks at a protein list. Three choices set the ceiling on what the data can support: whether the question is discovery or targeted at all, whether the mass spectrometer acquires by data-dependent acquisition (DDA) or data-independent acquisition (DIA), and how the sample was digested and cleaned up before it ever reached the column. Everything downstream — depth, missing values, quantitative precision, whether a fold change survives replication — is downstream of those three.

This guide is written as those decisions, in the order you actually make them, followed by the quality-control metrics that tell you whether a finished run is usable. It is deliberately not a primer on how a mass spectrometer works; if you need the upstream instrumentation, start with how liquid chromatography and mass spectrometry are coupled and electrospray source tuning, then come back.

Decision 0: discovery or targeted? (and why this is not the same instrument)

The most common wasted proteomics experiment is a discovery run asked to answer a targeted question. The two acquisition families are built for different jobs and generally on different hardware.

Discovery (DDA and DIA) asks what is in this sample and what changed? It runs on a high-resolution, accurate-mass analyser — an Orbitrap or a high-resolution time-of-flight — because identification depends on resolving and mass-measuring peptides you did not specify in advance. It is hypothesis-generating and quantifies on a relative scale across the runs in the experiment.
Targeted (MRM/SRM and PRM) asks how much of this specific peptide is present? Multiple reaction monitoring runs on a triple quadrupole with both quadrupoles parked on pre-declared precursor and product masses. Nothing you did not list is measured, which is precisely why sensitivity, dynamic range and quantitative accuracy are better for the things you did list. Parallel reaction monitoring (PRM) is the high-resolution cousin: a targeted precursor list on an Orbitrap or QTOF, with the full fragment spectrum recorded rather than selected transitions.

The practical rule: use discovery to generate a candidate list, use targeted acquisition to verify it in more samples with real quantitative rigour. Running DDA on 200 clinical samples to measure four proteins is an expensive way to get a worse answer than an MRM assay would have given. The architecture trade-off is covered further in QTOF vs triple quadrupole.

Decision 1: DDA or DIA

Both start the same way — a bottom-up (shotgun) workflow in which proteins are digested to peptides, separated by reversed-phase LC, ionised by electrospray, and fragmented. They differ entirely in how the instrument decides what to fragment.

How DDA acquires

The instrument records a survey MS1 scan, ranks the precursor ions in it by intensity, and selects the top N for isolation and fragmentation as individual MS2 scans. A dynamic exclusion window then suppresses each selected precursor for a set period so the next cycle moves down the abundance list rather than re-sampling the same peaks.

Two consequences follow directly and they are the whole argument against DDA:

Precursor selection is stochastic near the selection boundary. A peptide that ranks just inside the top N in one run ranks just outside it in the next. The result is missing values — a peptide quantified in six of ten runs, absent in four, for reasons that have nothing to do with biology. This is the single largest source of attrition in DDA datasets and it worsens as sample number grows.
Depth is intensity-biased. Top-N selection systematically favours abundant peptides, so low-abundance proteins are under-sampled unless you buy depth back with fractionation or longer gradients.

DDA’s compensating strengths are real: each MS2 spectrum comes from (nominally) one precursor, so spectra are clean and interpretable, database searching is mature and well understood, and the workflow needs no prior spectral library. Software-level rescue for missing values exists — match-between-runs style transfer of identifications across runs by accurate mass and retention time — but it is an inference, not a measurement, and it should be reported as such.

How DIA acquires

DIA abandons precursor selection. The instrument steps through a series of m/z isolation windows spanning the precursor range and fragments everything in each window, cycling continuously. Nothing is decided at run time, so nothing is decided differently between runs. SWATH-MS is the best-known named implementation of this idea, originally described on a QTOF using a series of sequential wide isolation windows across the precursor range; contemporary methods almost always use variable-width windows, narrow where the precursor density is high and wide where it is sparse, rather than a uniform width.

The trade-off moves from acquisition to data analysis. Every MS2 spectrum is chimeric — fragments from every co-isolated peptide in that window, superimposed. Extracting individual peptides from that requires either:

a spectral library, typically built from DDA runs of the same or a comparable sample (increasingly from fractionated pooled samples, or predicted in silico from sequence); or
a library-free search, in which the tool predicts spectra and retention times from a protein FASTA and searches the DIA data directly. This is now the default in several widely used pipelines and has largely removed the “you must build a library first” objection to DIA.

What you get for it is data completeness: because every precursor in range was fragmented in every run, quantitative values are present for far more peptides across far more samples, and the missing-value problem shrinks dramatically. That, not raw identification count, is DIA’s real advantage — and it is worth most exactly where DDA hurts most, in large cohorts.

The selection rule

If this describes your experiment	Choose	Why
Large sample count (tens to hundreds), quantitative comparison across all of them	DIA	Missing values compound with sample number; DIA is the only one of the two that does not degrade as n grows.
Small number of samples, maximum possible catalogue depth, offline fractionation available	DDA	Fractionated DDA still reaches very deep catalogues, and spectral cleanliness helps with unexpected identifications.
Post-translational modification discovery with unanticipated or open modification searching	DDA	Open/unrestricted modification searching is far better established against clean, single-precursor MS2 spectra.
Low-input material — single cells, laser-capture microdissection, needle biopsies	DIA	No precursor intensity threshold has to be cleared for a peptide to be fragmented, so sampling does not collapse when signal is scarce.
You need isobaric multiplexing (TMT) in the same run	DDA	Isobaric reporter-ion quantitation is a DDA-native workflow; combining it with DIA is possible but far less standard.
Cohort will be extended later with additional batches	DIA	Batch-independent acquisition; new runs are quantified against the same library or FASTA rather than being tied to a labelling batch.
You already know the ten proteins you care about and need absolute quantitation	Neither — go targeted	See Decision 0. MRM or PRM with stable-isotope-labelled internal standards.

One thing that is not a valid selection criterion: a published identification count. Protein-group counts depend on the sample, the gradient length, the instrument generation, the FASTA, the FDR level and the inference rules used to collapse peptides into proteins. Comparing your run against a number from a paper that used different settings tells you nothing.

Decision 2: the quantitation strategy

Acquisition mode and quantitation strategy are separate choices that are frequently conflated. There are three practical families.

Label-free quantitation (LFQ)

Each sample is prepared and run separately; quantitation comes from MS1 precursor intensity (extracted ion chromatogram area) or, in older work, spectral counting. Cheapest per sample, unlimited sample number, no labelling chemistry to fail. The cost is that every source of run-to-run variability — digestion efficiency, loading amount, column condition, source stability — lands directly in the quantitative result, so normalisation and replication carry more weight. LFQ is the natural partner for DIA and is used widely with DDA too.

Isobaric labelling (TMT / iTRAQ)

Peptides from several samples are labelled with tags of identical total mass but different isotope distribution, then pooled and run as a single sample. The tags are indistinguishable at MS1 and split into distinct reporter ions in the low-mass region of the MS2 spectrum, whose relative intensities give the ratios between channels. TMT has shipped in 6-, 10- and 11-plex forms and TMTpro in higher-plex forms; confirm the current channel count against the supplier’s specification before designing a batch layout, since it has increased across reagent generations.

Benefits: sample-to-sample variation in LC and source performance cancels, because all channels experience the same run; instrument time per sample drops by the plex factor; missing values within a plex are almost eliminated.

The failure mode you must design around is ratio compression. Because DDA isolation windows are not infinitely narrow, co-eluting near-isobaric peptides are isolated alongside the target and their reporter ions add to every channel. Real fold changes are pulled toward 1:1 — sometimes severely. The established mitigations, in rough order of effectiveness and cost:

SPS-MS3 — select several MS2 fragments and fragment again, reading reporter ions from the MS3. Substantially reduces interference at the cost of instrument time and sensitivity, and requires an instrument that supports it.
Extensive offline fractionation — fewer co-eluting peptides per window.
Narrower isolation windows and ion-mobility separation (e.g. FAIMS) to reduce co-isolation before it happens.
Reporting the co-isolation / precursor purity metric your search engine produces and filtering on it. This one is free and routinely skipped.

The second design constraint is batching. Samples in different plexes are not directly comparable without a bridge, so every plex needs a shared reference channel (a pooled internal standard) and your batch layout must not confound biological groups with plex membership. A plex containing all the controls and another containing all the cases is an unrecoverable design error, not a normalisation problem.

Metabolic labelling (SILAC)

Heavy amino acids are incorporated during cell growth, so labelling happens before lysis and every subsequent handling step is shared between conditions. That gives it the lowest technical variance of the three. Its limits are structural: it needs cells you can culture through several doublings, and it does not extend to tissue, plasma or clinical material in the general case.

Decision 3: the sample prep choices that actually drive depth

More proteomics experiments are limited by the bench work than by the instrument. Each step below is a decision with a specific downstream consequence, and each is worth more attention than an extra hour of gradient.

Lysis and detergent removal

Strong detergents extract membrane and hard-to-solubilise proteins well — and are catastrophic downstream. SDS in particular suppresses electrospray, contaminates the column and persists for injections afterwards. So the choice is not “detergent or no detergent”; it is which detergent-removal method you commit to:

SP3 (single-pot, solid-phase-enhanced sample preparation) — proteins are aggregated onto carboxylate-coated magnetic beads, washed free of detergent and salts, and digested on-bead. Tolerant of low input, works in a single tube, easily automated.
S-Trap (suspension trapping) — protein is trapped as a fine particulate suspension in a quartz filter, washed and digested in place. Fast and tolerant of high SDS concentrations.
FASP (filter-aided sample preparation) — buffer exchange and digestion on a molecular-weight cut-off filter. The long-established method; slower and more prone to filter-dependent losses than the two above.
Acetone or chloroform/methanol precipitation — cheap and effective at removing detergent, with variable and sample-dependent recovery, particularly at low input.
MS-compatible surfactants — acid-cleavable detergents that are degraded before injection, avoiding the removal step entirely at higher reagent cost.

At low input the deciding factor is nearly always surface losses, not chemistry: peptides and proteins adsorb to tubes and tips. Single-tube workflows and low-bind consumables matter more than they sound like they should.

Reduction and alkylation

Disulfides are reduced (DTT or TCEP) and free cysteines capped, conventionally with iodoacetamide, giving the carbamidomethyl modification your search must be configured to expect as a fixed modification. Two failure modes: incomplete alkylation leaves a mixed population that splits each cysteine-containing peptide’s signal across two forms, and over-alkylation — too much reagent, too long, or light exposure — adds the same mass to methionine, histidine and peptide N-termini, generating unassignable spectra and depressing identification rate. Iodoacetamide reactions are run in the dark and stopped; both instructions exist for this reason.

Digestion

Trypsin is the default protease because it cleaves C-terminal to lysine and arginine, and is inhibited when the following residue is proline. That produces peptides in a mass range and with a C-terminal basic residue that suit electrospray and collision-induced fragmentation well. The decisions that matter:

Protease choice. Trypsin alone leaves parts of the proteome invisible — regions with no cleavage sites in reach produce peptides too long or too short to detect. Lys-C is commonly used ahead of trypsin (Lys-C tolerates higher urea concentrations, improving cleavage in denaturant); Glu-C, chymotrypsin, Asp-N and Arg-C are used to reach sequence space trypsin cannot, at the cost of extra runs. For PTM site localisation in a specific region, protease choice is often the whole experiment.
Enzyme-to-protein ratio and time. Under-digestion shows up as a rising missed-cleavage rate, which both loses signal (one peptide’s intensity split across several forms) and distorts label-free quantitation. Over-long digestion increases non-specific cleavage and autolysis products.
Denaturant compatibility. Urea concentrations that keep protein unfolded also inhibit trypsin; the usual resolution is a Lys-C step at high urea followed by dilution before trypsin, and urea must be kept below temperatures at which carbamylation becomes significant.

Cleanup before injection

Digests are desalted by reversed-phase solid-phase extraction — C18 tips, StageTips or cartridges — before injection. This is a real method-development step, not a formality: sorbent chemistry, wash strength and elution strength determine whether hydrophilic peptides are lost in the wash and whether hydrophobic ones ever come off. The same selection logic applies here as anywhere else in SPE; see SPE cartridge selection and the wash-step trade-off for how to reason about it. Keratin contamination — skin, dust, wool — is introduced at exactly this stage and is the most common contaminant in a disappointing spectrum count.

Fractionation: the main lever on depth

Nothing else buys catalogue depth as reliably as reducing sample complexity before the analytical column. High-pH reversed-phase offline fractionation is the current default because it is genuinely orthogonal to the low-pH reversed-phase separation used online, while using compatible solvents; strong cation exchange and gel-based fractionation (in-gel digestion of excised bands, following SDS-PAGE) remain in use. Size exclusion is used at the protein level for complex-level questions rather than for depth.

The trade-off is linear and unavoidable: n fractions means n times the instrument time per sample. That is why fractionation pairs naturally with small-n, deep DDA experiments and with TMT (where the plex amortises the cost across channels), and rarely with large-cohort DIA.

Enrichment for PTMs

Modified peptides are a small fraction of the digest and are not found by looking harder. Phosphopeptides are enriched by immobilised metal affinity chromatography or titanium dioxide; ubiquitination remnants and acetyl-lysine by antibody-based enrichment; glycopeptides by lectin affinity or hydrophilic interaction chromatography. Enrichment specificity — the fraction of identified peptides that actually carry the modification — is itself a QC metric worth recording per batch.

The QC metrics that say a run is usable

Almost every threshold below is instrument-, method- and lab-specific. A number that indicates a healthy Orbitrap on a two-hour gradient with a HeLa digest standard means nothing on a different platform, gradient or matrix. The correct use of this table is to establish your own baseline on your own system-suitability sample and then watch for drift. Absolute numbers copied from a paper are not a QC criterion.

Metric	What it actually diagnoses	How to read it
Median precursor mass error (ppm)	Calibration state of the analyser	Should sit close to zero and tight around it. A systematic offset means recalibrate; a widening distribution across a run points at space-charge or thermal drift.
MS2 identification rate (PSMs at 1% FDR ÷ MS2 scans acquired)	Whether the instrument is fragmenting things worth fragmenting	The single most informative composite metric. A drop with unchanged sample points at the source, the column or the digest; a drop with a new sample type may be biology.
Peptide and protein-group counts	Overall system performance	Only interpretable against your own historical baseline for the same standard, gradient and FDR settings.
Chromatographic peak width (FWHM)	Column and connection health	Broadening across a sequence is the classic early sign of column degradation, a void, or a dead volume in a refitted connection.
Retention-time reproducibility of spiked standards (iRT / PRTC peptides)	LC reproducibility, and the validity of RT-based matching	Directly gates match-between-runs and DIA library matching. Track the regression of observed RT against the standard’s indexed values, not just the raw drift.
Missed-cleavage rate	Digestion completeness	A rise batch-over-batch means enzyme activity, ratio, time, or denaturant carry-over — a bench problem, not an instrument problem.
Oxidised-methionine percentage	Oxidative stress during sample handling	Elevated values point at sample age, sonication, or an oxidising source region rather than the instrument’s detection.
Charge-state distribution	Source condition and digestion	Excess singly-charged species or a shift in the 2+/3+ balance is a source or a chemistry signal, well before identification rate falls.
Ion injection / accumulation time and TIC stability	Loading, spray stability, source contamination	Rising injection times at constant load mean less signal reaching the analyser. Spray instability shows here first.
Precursor purity / co-isolation	Interference — critical for TMT	Filter reporter-ion quantitation on it. Skipping it is the most common cause of unexplained ratio compression.
CV of quantified proteins across technical replicates	Whether an observed fold change can be believed	Establish it once per workflow. It sets the effect size your experiment can actually resolve — which is the number a power calculation needs.
Carryover in blank injections	Cross-contamination between samples	Run blanks after high-load samples. Carryover in a cohort study is a confound, not a nuisance.

Run a system-suitability sample, on a schedule, and keep the results. A commercial standard digest (HeLa or similar) and a spiked retention-time peptide mixture such as PRTC or iRT peptides give you a stable reference that changes only when the system changes. Trending those results over months is what converts QC from a per-run judgement call into an early-warning system; open tooling exists for exactly this, including Panorama AutoQC for automated longitudinal tracking of Skyline-processed system-suitability runs.

What “identified” is actually claiming

Identification numbers are only comparable when the statistics behind them are stated, and three things routinely go unreported:

The FDR level. Target-decoy searching against a reversed or shuffled database gives an empirical false discovery rate, and 1% is the field’s convention — but 1% at the peptide-spectrum match level is not 1% at the protein level. Protein-level FDR is looser than PSM-level FDR at the same nominal cutoff, and the gap grows with database size and sample complexity. State which level you filtered at.
Protein inference. Bottom-up proteomics measures peptides and infers proteins. Peptides shared between homologous proteins or isoforms cannot distinguish them, so search engines report protein groups and assign shared peptides by rules (the “razor peptide” convention being the best known). A protein-group count is not a count of distinct protein species, and single-peptide identifications deserve separate treatment.
What was inferred rather than measured. Match-between-runs and library-based DIA extraction both propagate identifications by mass and retention time. That is legitimate and useful; presenting it without distinguishing it from directly acquired identifications is not.

Standards, formats and deposition — the part that outlives the run

Proteomics has an unusually mature open-standards ecosystem, maintained by the HUPO Proteomics Standards Initiative (HUPO-PSI). Using these formats is what makes a dataset reusable rather than a folder of vendor binaries, and journal and funder expectations increasingly assume them.

mzML — the open format for raw spectrometer output. Version 1.1.0 was released on 1 June 2009 and the PSI describes it as stable ever since; it superseded the earlier mzData and mzXML formats, which are deprecated. Note for DIA specifically: a 1.2 schema revision was once planned to add ion-mobility and DIA support, but the PSI’s position is that both can be encoded without a schema change, using additional controlled-vocabulary terms. A proposed best-practices document for encoding IM-MS and DIA in mzML is out for community review (version 1.0.0rc, the PSI’s mzML page updated 11 August 2026) and is partly implemented in ProteoWizard/MSConvert. If you are archiving DIA data, follow that proposal rather than inventing local conventions.
mzIdentML — identification results: what was searched, with what parameters, and what matched.
mzTab — a simpler tab-separated report format for final identification and quantitation results.
mzQC — the quality-control exchange format, reached final release as version 1.0.0 on 21 February 2024. It is designed to serve as handover, report and archive format for exactly the metrics in the table above, which is what makes longitudinal QC portable between tools rather than locked in one vendor’s dashboard.
mzSpecLib and the Universal Spectrum Identifier (USI) — a standard spectral-library format and a standard way to reference one specific spectrum in one specific public dataset, so a claim about a spectrum can be checked rather than taken on trust.
SDRF-Proteomics — sample-and-data-relationship metadata: which run corresponds to which sample, condition, replicate and label channel. This is the file that decides whether a reanalyst can reconstruct your design at all.
TraML — transition lists, the bridge to targeted assays.

The PSI also maintains the controlled vocabularies these formats depend on, and additional specifications including PEFF, ProForma, mzPAF, proBAM/proBed and FeatureTab. The full current list is published on the PSI’s own specifications pages, which is where to check before assuming a format is current.

Deposition

Raw data deposition is now the normal expectation for published proteomics. Submissions go to a ProteomeXchange member repository — PRIDE at EMBL-EBI, PeptideAtlas, MassIVE, jPOST, iProX or Panorama Public — and receive a PXD accession that goes in the manuscript. ProteomeXchange is the coordinating consortium and index, not a repository itself; ProteomeCentral is the unified search across members. Submissions are categorised as “complete” (fully annotated with identification results) or “partial”. Molecular & Cellular Proteomics has required raw data deposition with submitted manuscripts since 2015, and other publishers have moved in the same direction; check the specific journal’s current author guidelines rather than relying on a general statement.

Two practical points that connect this to the rest of your data governance. First, a domain repository deposit is normally what a funder’s data management plan should name for proteomics data — a generalist repository is a weaker answer when a discipline-specific one with format validation exists. Second, the PSI format stack is precisely what operationalises FAIR data principles here: mzML makes the data accessible in a non-proprietary form, the controlled vocabularies make it interoperable, and SDRF-Proteomics plus a PXD accession make it findable and reusable. Protein identifiers should resolve to a stable reference — see UniProt — and the exact FASTA version searched should be recorded, because identification results are not reproducible without it.

Troubleshooting: symptom to cause

Symptom	Look here first
Identification rate down, mass accuracy fine, peak shape fine	Digestion or cleanup. Check missed-cleavage rate and oxidation percentage before touching the instrument.
Peaks broadening across a sequence	Column bed degradation, a void, or dead volume in a recently remade connection.
Unstable spray, erratic TIC	Emitter tip, source contamination, or bubbles/leaks in the LC — see source parameter tuning.
Many spectra, few identifications, unassignable masses	Over-alkylation, unexpected modifications, wrong or outdated FASTA, or polymer/detergent contamination.
TMT fold changes all near 1:1	Ratio compression. Check precursor purity, then isolation width, fractionation depth and whether MS3 is available.
DDA cohort riddled with missing values	Structural, not fixable by re-running. Either accept imputation with the assumption stated, or move the cohort to DIA.
DIA quantifies far fewer proteins than expected	Library or FASTA mismatch, retention-time calibration failure, or isolation windows badly matched to the precursor density of this sample.
Contaminant proteins dominate	Keratin from handling at the cleanup stage; trypsin autolysis peptides; serum albumin carry-over from a previous high-abundance sample.
Batch effects align with plex membership	Design error. A pooled bridge channel in every plex is the fix, and it has to be planned before labelling.

Frequently asked questions

What is the difference between DDA and DIA in one sentence?

DDA lets the instrument choose which precursors to fragment at run time based on intensity, which makes clean spectra but different choices in every run; DIA fragments everything within stepped m/z windows regardless of intensity, which makes chimeric spectra but the same measurement in every run.

Is DIA always better than DDA?

No. DIA is better where run-to-run completeness dominates — large cohorts, low input, longitudinal studies. DDA remains preferable for open or unrestricted modification searching, for isobaric-labelled multiplexed designs, and for very deep catalogue building on a small number of heavily fractionated samples.

How is DIA different from MRM?

Both avoid intensity-driven precursor selection, but for opposite reasons. MRM measures a short, pre-declared list of transitions with maximum sensitivity and quantitative accuracy on a triple quadrupole; DIA measures everything in a mass range on a high-resolution instrument and resolves it computationally afterwards. MRM is targeted verification; DIA is discovery that happens to quantify reproducibly. They are complementary, and the usual sequence is DIA or DDA first, MRM second.

Do I still need a spectral library for DIA?

Not necessarily. Library-free (direct) DIA analysis, where the tool predicts fragment spectra and retention times from a protein sequence database, is now well established and is the default in several pipelines. An experimental library built from fractionated DDA runs of the same matrix can still add depth, particularly for unusual sample types, but it is no longer a prerequisite.

What FDR should I report and at what level?

One percent is the field’s convention, but the level matters more than the number. Report explicitly whether 1% was applied at the peptide-spectrum match, peptide or protein level — they are not equivalent, and protein-level control at a nominal 1% PSM FDR is looser than it appears, especially with large databases.

Why does my protein count differ so much from a published study?

Because protein-group counts are a function of sample, gradient length, instrument generation, fractionation depth, FASTA content and version, search engine, FDR level and protein-inference rules. Two of those differing is enough to change the number substantially. Compare against your own baseline on your own standard, not against a paper.

Should I use TMT or label-free?

TMT when you have a fixed, moderate number of samples that fit cleanly into whole plexes with a pooled bridge channel, want to eliminate within-plex missing values, and can afford the mitigation for ratio compression. Label-free when sample number is large or open-ended, when true fold-change magnitude matters more than precision, or when the labelling chemistry is a risk you do not want in the workflow.

How much protein do I need to start?

It depends entirely on the workflow, and any single number quoted without that context is misleading — modern low-input methods work far below what a conventional fractionated workflow assumes. The more useful framing: determine the minimum input at which your prep gives acceptable replicate CVs, by titrating it once, and treat that as your floor. Below it, losses to surfaces rather than instrument sensitivity are usually what limits you.

What QC metric should I check first when a run looks wrong?

Median precursor mass error and MS2 identification rate together. Mass error isolates whether the analyser is calibrated; identification rate at correct mass accuracy points the investigation at the chromatography or the sample chemistry instead of the instrument.

Which file formats should I archive?

Convert vendor raw files to mzML for the spectra, keep mzIdentML or mzTab for results, record SDRF-Proteomics for the sample-to-run mapping, and deposit to a ProteomeXchange member repository for a PXD accession. Archive the exact FASTA and search parameters alongside them; without those, the results are not reproducible even from perfectly preserved raw data.