What is new in HDX Workbench, written for the people who run the experiments rather than the people who wrote the code.
A complete data-independent acquisition workflow, from building an undeuterated fragment library to reading deuterium uptake off individual fragments.
Import a PFNet model and it sits beside your measured data on the same colour scales, with the residues it could not resolve marked rather than smoothed over.
A volcano plot that applies the accepted hybrid test, so a peptide counts as changed only when the change is both large enough and reproducible enough.
Data-independent acquisition was the single largest piece of work in this period. Fragments do two jobs: they localise a peptide's uptake more finely than the peptide itself, and they show independently when that peptide actually eluted. The second is what removes most of the manual retention time curation from a run, and lets analysis proceed with far less intervention.
Files that need converting to mzML are converted for you. If the required OpenMS tools are not installed, or the data is not actually DIA, the run stops immediately with a dialog that says so rather than quietly producing nothing.
Where Peptide set right-click, DIA DetectA checkbox skips rebuilding the undeuterated library when you already have one, which saves the longest part of a repeat run.
A cross-cell view: the fragment table across the top and a deuteration scatter below it, for spotting which cells in the whole experiment disagree before you go looking at individual spectra.
Where Spectra window, DIA Quality tabSelect a single replicate and you get four stacked panels for it: the precursor chromatogram, one chromatogram per detected fragment shaded by whether the fragment agreed with the others, the labelled fragment spectrum, and the deuteration model fitted through them. A properties sidebar carries the peptide, deuteration and quality numbers.
Where Spectra window, MS2 tab, with exactly one replicate selected
A panel for judging confidence across the run: confidence tiers, counts of cleanly matched fragments, whether the charge states agree with each other, and whether the fragment retention time matched the precursor. It also reports how far the DIA evidence would move each retention window if you let it.
Fragment uptake is fitted as a slope across the fragment series and normalised to the precursor, so a single noisy fragment no longer swings the answer.
Most bad HDX numbers come from measuring the right peptide at the wrong moment. Several features were added specifically to find, check and correct retention windows.
Runs a Sage search over your undeuterated files and updates the peptide set's retention times from the identifications. It offers a dry run first, so you can see what would change before anything is written, and it explains itself and stops if Sage is missing or your files carry no fragment spectra.
Where Protein right-click, Refresh RTs from MS2 (Sage)Where fragment evidence says a peptide elutes somewhere other than the saved window, the correction can be previewed as a table before being applied, and applying it can be rolled back completely.
A window is only corrected when other charge states of the same peptide, or the other replicates of the same cell, agree with the new position. A single outlying measurement cannot drag the window on its own.
Peptide sets are checked against the retention time the peptide was originally identified at, and cells that landed a long way off are flagged for you to look at.
Detection now searches the full declared retention window for the peak apex rather than a narrow slice of it, with score gates so a wider search does not mean a worse pick. Windows that had to be rescued this way are flagged rather than passed off as ordinary picks.
A group of changes aimed at one failure mode: the software confidently reporting deuteration for a peak that is not your peptide.
Detection now checks for a peak sitting just before the peptide's first isotope, for evenly spaced peaks that suggest a different charge state overlapping, and for peptides whose masses conflict with each other. Any of these can veto a pick that would previously have been accepted on score alone.
A parts-per-million ceiling is applied to the combined undeuterated spectrum, so a peak at approximately the right mass is no longer good enough.
The window used to quantify a peptide is capped at the mass it could reach if every exchangeable hydrogen were deuterated. This also fixed a case where the reported centroid could fall outside the very window it was measured in.
Cells that produced nothing on the first pass can be rescued from the combined spectrum, and cells whose signal only appears in a single scan can be latched onto when the evidence supports it. Every rescue is recorded on the cell rather than hidden.
A replicate that was marked usable but produced no measurement used to enter the averages as a placeholder value, quietly pulling means and standard deviations off. Those replicates are now left out and counted separately.
Support for peptides whose isotope envelope splits into two populations, the signature of EX1 exchange or of two conformations in solution.
A toggle in the spectra view fits two populations to the displayed envelope and draws both, with each one's deuteration, share of the population and how much better the two-population fit is than a single one. You choose which population to assign, and assigning one the data does not support asks you to confirm first.
The assignment is saved with the peptide, appears in the results grid, and is written to the CSV export.
Where Spectra window, bottom bar of the signal view
PFNet models exchange at individual residues. HDX Workbench can now bring those results back in and show them next to your measured data.
Import the zipped output and it is filed under the protein. Re-importing replaces that set and leaves any others alongside it, so several models can coexist.
Where PFNet menu, Import PFNet Results (Zip)The coverage dropdown gains a consolidated view per state and a state-versus-control view. Under each stretch of sequence sit three aligned tracks: your measured uptake averaged onto residues, the model's residue-level prediction on the same colour scale, and the model's confidence. Residues the model is least sure of are hatched rather than filled.
Because all three share the sequence ruler, you can see where the model reproduces what you measured, where it departs from it, and whether it was confident when it did. Both views export residue-level CSV files and PyMOL scripts for structure figures.
Per-residue exchange rates or free energies as bars with uncertainty, a track showing which stretches the model could not resolve to single residues, and low-confidence residues greyed so a prediction is never mistaken for a measurement.
Where Spectra window, PFNet tabWild type against a point mutant, from one protein entry and one peptide set. The substitution is declared on the wild-type peptide, and both variants are then measured side by side in the same run.
A tagged row is not a mutant row. It defines the wild-type and the mutant candidate together, so there is no second row to add for that peptide and charge. Give the pair a retention window wide enough for both, since a substitution can shift when the peptide elutes.
Where Input peptide set CSV, sequence columnThe number in a tag is a one-based position within that peptide, not a position in the protein or in a PDB file. One protein mutation therefore takes a different number in every peptide that covers it, and each charge state of an affected peptide carries its own tag.
Substitutions are joined inside a single tag with hyphens. AASATAA[MUT-3Y-5P] derives AAYAPAA, with both changes belonging to the same mutant variant rather than describing two separate mutants.
Five columns, with explicit retention-time bounds. The example below carries all three rules at once: both charge states of the tagged peptide are tagged, the second peptide has the position recounted from its own start, and the peptide that does not cover the mutation is left plain.
rt_start,rt_end,charge,sequence,ms score 6.20,6.80,2,AASAA[MUT-3Y],1.0 6.20,6.80,3,AASAA[MUT-3Y],1.0 6.10,6.90,2,GAASAA[MUT-4Y],1.0 8.00,8.60,2,AAAAA,1.0
Retention times are in minutes. Keep the header and the column order, use uppercase one-letter residue codes, and leave no spaces inside a tag. This is an input peptide set rather than a detected-results export, so exported molecular formulas and masses are not something to edit by hand.
Where Saved as a comma-separated .csvIn the Perturbation view the two variants share a comparison row while keeping their own sample observations, so the pair reads as a comparison without hiding what was actually detected in each sample.
Where Perturbation viewIf both variants are marked valid in the same sample at the same timepoint, statistics for that cell are withheld instead of averaging two populations into one number. The detections stay on screen for you to inspect, and an incorrect one should be rejected only after looking at the evidence behind it.
Deciding which differences between two states are real, using the method the field has settled on.
A t-test alone will flag a tiny difference on a very reproducible peptide. A fixed cutoff alone will flag a large difference on a noisy one. The hybrid test requires both at once: the difference must be statistically significant and larger than a threshold calculated from how reproducible your own replicates actually were. Everything else stays on the plot, greyed, with a note saying which of the two gates it failed.
One point per peptide per labelling time, with the size of the difference across the plot and the strength of the evidence up it. Colour says what the point is: protected, deprotected, significant but small, large but noisy, or nothing detectable.
Both cutoff lines can be dragged. Hovering a point gives the full numbers behind it. Clicking one selects that peptide everywhere else in HDX Workbench, so its spectra and uptake curve come up. Dragging a box around a group selects them together for marking or export.
Where Spectra window, Volcano tab, and the Volcano menu
The plot exports as an image, the points or just the hits export as CSV, and a settings snapshot records the comparison, the replicates used, the threshold and how it was calculated, and the software version. The full CSV export gains the same verdict columns.
The sequence coverage map now greys non-significant peptides using the same test as the volcano plot, so two views can no longer disagree about what counts as a change.
A standardised file for HX/MS results, written alongside the usual CSV. It carries the isotope envelope for every measurement, not only the centroid, which is what lets a downstream model work from the peak shape.
Save Report writes a .hxms file for each state into the export folder, next to the CSV and HDXer files. The output is byte-identical to what the PFLink converter returns for the same CSV, so the files can be handed to PFNet or FEATHER without a conversion step in between.
Where Save Report, alongside the CSV exportIllustrative, with the field order the format defines:
METADATA PROTEIN_STATE apo METADATA TEMPERATURE(K) 298.15 METADATA pH(READ) 7.00 REMARK HXMS_DATA_FORMAT v1.0 TITLE_TP INDEX MOD START END REP PTM_ID TIME(Sec) UPTAKE ENVELOPE TP 1 A 40 44 0 0000 3.0e+01 2.41 0.02,0.11,0.28,0.31,0.19,0.07,0.02 TP 2 A 40 44 0 0000 inf 4.86 0.01,0.06,0.19,0.33,0.26,0.11,0.04
Uptake is the centroid mass minus the peptide's own zero-second centroid. The fully deuterated control is written with a time of inf rather than a large number, so it cannot be mistaken for a very long labelling point.
The field is filled from the isotope distribution the Workbench already measured, so a residue-level model is not restricted to the centroid and can be run on the peak shape instead. This is what connects the export to PFNet's envelope model.
Every charge state of a cell gets its own record, since charge is not part of a peptide's identity in this format. Discarded replicates and the per-peptide aggregate rows are not exported. The zero-second reference is a replicate's own mean where it has one, and falls back to the other replicates when it does not.
Modification and mutation tags written in the peptide set become entries in the dictionary at the end of the file, so a downstream reader can resolve what a record refers to without going back to the peptide set.
When a fully deuterated control is present, HDX Workbench calculates how much deuterium was actually recovered and how much was lost to back-exchange, per peptide, and reports a quality rating for the control itself.
A recovery column in the results grid, so peptides with poor deuterium recovery are visible while you review rather than only in the exported file.
Several runs that used to be left overnight now finish while you wait. The results were checked to be identical before and after each change.
Chromatogram extraction was rewritten to read each file once, over only the retention range actually needed, and to pull all isotope channels in a single pass.
The same windowing applied across the readers, plus opening each raw file once per peptide instead of repeatedly, and searching within spectra rather than scanning every peak.
Statistics are calculated only for the columns you are looking at, the project tree loads a branch when you expand it, and the intensity plots build in the background instead of blocking the window.
One menu item runs the undeuterated detection, writes the peptide set from everything it found, and runs the HDX job from it, reusing the undeuterated results instead of measuring them twice. If any step fails, it cleans up after itself.
Where Protein right-click, Detect + HDX (one pass, reuse 0s results)Each detection run writes a timing breakdown next to the job, so a slow run can be attributed to a stage rather than guessed at.
Three ways to extend an experiment without reprocessing it: one more peptide, a list of them, or a whole new timepoint. Each reuses the settings the run was made with, and only the new work is searched.
Right-click the HDX job and add the peptide there, so it inherits that job’s working peptide set and the parameters it was run with rather than a fresh set of defaults. The sequence and charge are validated before anything starts, only the added peptide is searched, the results are saved the way a normal run saves them, and the view refreshes so the peptide appears alongside the rest.
Where HDX job right-click, Add peptide(s)The same column-mapping interface the protein editor already uses, so there is nothing new to learn about matching your columns. Sequence and charge are required and retention time is optional. Duplicate and invalid rows are reported rather than dropped quietly, and an import that stopped part way can be resumed from what is left.
Where Add peptide(s), CSV importDrop the raw files into a box for each sample and replicate. The timepoint is registered against the default HDX job’s full peptide set, and only the new files are searched, so the timepoints already measured are left exactly as they were.
Where HDX job right-click, Add timepointTriangle handles on the chromatogram and the spectrum set the window boundaries directly. The text boxes follow as you drag, and pressing Enter afterwards applies the change. The old mass range slider is gone. Zooming still works; the handles only take over while you are actually holding one.
On a multi-cell view the boxes start empty and each cell keeps its own boundary, so you can move one edge across every loaded cell without flattening the other.
One level of undo for the Set button, restoring every field that the recalculation changed.
Matching the convention used in the field, and the same scale the coverage maps use.
The Marked and Enabled columns are checkboxes you click, instead of a right-click menu, and the change is saved as you go.
Expanded and collapsed branches survive closing and reopening a project, and expanding no longer takes two clicks.
Conversion and search progress is reported instead of the window appearing to hang.
A corrupt result file is skipped with a message rather than breaking the whole load, and there is a repair action that regenerates the affected files.
Sample assignments are validated before the experiment is created, so mistakes surface while you can still fix them easily.
A command-line tool builds experiments from a CSV description, for setting up many experiments at once or scripting a reprocessing run.
Column widths were retuned, the default sort order fixed, and the peptide features column is hidden when nothing populates it.
The three major vendors are now handled by one common reading layer, which means a feature added for one instrument tends to arrive for the others too.
Bruker folders are recognised throughout the application: in the experiment wizard, when browsing to a folder, in the project tree, and in the spectra views. Ion mobility is taken into account when spectra are combined.
By default the raw data stays where it is and the project keeps a pointer to it, so a large acquisition is not duplicated into the project folder. There is a checkbox if you would rather copy it, and it checks you have the disk space first.
Where Experiment wizard, raw data stepThe vendor detected from your files is shown while you set the experiment up, and a batch that mixes vendors is refused with an explanation instead of failing later in the run.
Lockmass correction is applied when spectra are read, and profile spectra are converted to peaks at the reading boundary rather than later, which removes a source of small mass errors. Detection defaults were retuned for Waters instruments.
Where Detect dialog defaults, and automatically during any Waters runThe instrument's real isolation window boundaries are read out of the raw file instead of being inferred, which matters for assigning fragments in DIA runs.