Data-dependent and data-independent acquisition

Data-dependent and data-independent acquisition are two ways to spend the same scarce resource: the time the mass spectrometer has to fragment peptides while they are eluting. Neither way is universally the better purchase. Bottom-up proteomics in plain language describes what a fragment spectrum is for. This page is a selection guide for the sampling logic you write into a specification.

How a data-dependent method spends fragment time

A data-dependent run starts with a survey scan that lists precursor masses and intensities. The instrument then picks a short list, often the tallest peaks that are not on an exclusion list, and fragments them one after another. Dynamic exclusion tries not to fragment the same precursor again for a few seconds, so the list can move on to the next peak. The spectra that come back are relatively pure, which is why classical search engines are comfortable with them.

The cost is stochastic sampling. A peptide that is just inside the list in one run can fall just outside it in the next, because the spray flickered, a neighbour became taller, or the exclusion window landed differently. The peptide may still be in the vial. The fragment spectrum is what went missing. Discovery lists therefore contain holes that look biological if you forget the sampling. Sample preparation still sets the ceiling: a digest full of detergent or dominated by a few proteins gives the picker very little worth picking. Preparing peptides for mass spectrometry is the upstream control on that ceiling.

How a windowed method spends the same time

Data-independent acquisition fragments everything inside a mass window, then steps the window across the range. The spectra are mixtures. Several precursors contribute fragments at once. The gain is completeness: a peptide does not have to win a popularity contest to be fragmented. The same window scheme, repeated on every sample, is why a defined panel often compares more calmly across a long study.

The analysis is the new dependency. Software has to decide which fragments belong to which peptide. Many workflows match the windows to a spectral library recorded from earlier data-dependent runs of similar samples, or to a library predicted from a sequence database. Library-free approaches demultiplex directly against the database. All of them can be done well, and all of them can be done with an error rate nobody has looked at. The false discovery rate has to be a number you can explain for the peptide list and, separately, for the protein list. Proteomics Standards Initiative material is a public home for how such results are supposed to be described. It is not a certificate that a particular software default is conservative.

Choosing without crowning a winner

Use a careful data-dependent design when the point is to discover proteins you did not know to expect, and when the laboratory can search a stated database with a stated decoy strategy. The holes in the table are the price. Fill them, if the claim needs them, with a second acquisition on the same peptides, a targeted measurement, or a windowed method whose error control you trust. Do not fill them by relaxing the score until the missing protein appears.

Use data-independent acquisition when a defined set of proteins must be compared across many samples and the instrument and the software are both in place. The panel can come from a library you built, from a predicted library, or from a library-free search if the false discovery rate can be defended. A windowed run is a poor purchase when nobody in the facility can demultiplex it. The raw file will sit complete and unread.

Library-free data-independent analysis is sometimes offered as the way to discover the unexpected without a prior library. It can be that, on a method the laboratory has already pressure-tested. It can also be a confident list of demultiplexing mistakes. Ask who will defend the error rate. If the answer is a default checkbox, you have not selected a method yet. HUPO discussions of identification quality apply to both acquisition classes. A newer sampling scheme does not retire them.

There is a practical hybrid that specifications should name honestly. A data-dependent campaign on pooled or fractionated material can build a library, and data-independent runs can then quantify the samples. The library's breadth becomes the study's breadth. Proteins excluded from the library stay excluded. Write that limit into the plan so a later reader does not treat the quantified panel as the whole proteome.

Sampling, missing values, library, failure

PeptideAtlas is a public picture of which peptides the community has seen often enough to trust as recurring observations. It can suggest what a library might contain. It is not your sample's library. PRIDE is where both acquisition classes are deposited; the method section is the part worth reading before you copy a window width.

Failures that look like biology

A protein "found only in the treated group" in a data-dependent study may be a protein that won the picker only in those runs. Look at the survey scans and at a replicate prepared in a different order. If the precursor is present in the controls and was merely not fragmented, the claim shrinks to a sampling note.

A protein "quantified everywhere" in a windowed study may be a protein whose fragments were borrowed from a neighbour. The diagnostic is a manual look at the extracted fragment chromatograms: they should share a peak shape and a retention time. A match that is only a score in a table is not that look. When the laboratory cannot do that check, do not buy the method for a claim that will be challenged.

Sample preparation still creates empty files under either acquisition. A windowed method will faithfully fragment detergent polymers and keratin if that is what you injected. Acquisition choice does not wash a tube.

Research use, and what you are selecting

You are selecting a research acquisition method, not a diagnostic platform and not an instrument brand. The specification should be fulfillable on the instrument class the work will actually see. A paper's window scheme is not a requirement if the available analyser cannot cycle that fast or cannot resolve the fragments the software expects. Biosafety of the material is unchanged by how the ions are chosen. Institutional rules still decide containment. A beautiful sampling scheme does not make an infectious lysate into a routine chemical.

Writing the instrument you have into the request

Shared facilities differ. One site may have a mature windowed workflow, a maintained library and someone who can explain the error model. Another may have excellent data-dependent capacity and no demultiplexing practice. Specify the instrument class, the acquisition class, the library position (already built, to be built, predicted, or library-free), and the sample number. If the queue will be interrupted by other users, say so, because a data-dependent study split across a calibration change is two studies. Ask the method discussion to refuse a design the software stack cannot finish. The shotgun discovery proteomics reference, the protein identification by LC-MS/MS reference and the differential abundance reference are the pages that surround that conversation.

What the quote request should carry

Instrument class available, library or no library, sample number, discovery of unexpected proteins versus a defined panel, and the false discovery standard you expect to read in the report. Send those with the quote request. A method can be discussed from that list. "DIA if possible, otherwise DDA" is an acceptable preference only when you also say who will analyse the windows. Without that sentence you have asked for a file format, not for a result.