Article: Increasing mass spectrometry throughput using time-encoded sample multiplexing (Derks et al, 2025)

 

This work demonstrates an approach to multiplex samples for proteomic analysis using temporal encoding, which we refer to as 'timePlex'. This entailed the development of experimental implementations to stagger and overlap chromatography from multiple samples, and computational analyses to deconvolve the temporal-encoding as integrated as a module in JMod. Experiments were performed to benchmark proteomic coverage and quantitative accuracy of this new 'timePlex' data type to established methods (LF non-multiplexed DIA and plexDIA). Combinatorial multiplexing of 9-plexDIA and 3-timePlex enabled 27-plex data acquisition, achieving throughput of 500 samples/day with 25 minutes of active chromatography per sample.

 

Methods:

Bulk data:

Proteomics methods: plexDIA & timePlex

Single-cell data:

Proteomics methods: plexDIA & timePlex

Sample preparation method: nPOP
 

Data:

All raw and processed data from the article are organized in this MassIVE repository: MSV000097736.

 

Code:

JMod was used for searching all data. A step-by-step guide from raw files to search results is given below.

The {targets} R package was used to ensure repeatability of our downstream analyses. R scripts corresponding to benchmarking and single-cell analyses can be found in the "Data Analysis" folder of our GitHub. The raw data, processed data, libraries, and meta data required to repeat these analyses can be found at our MassIVE repository: MSV000097736

Code used to create RT prediction models for label-free (LF) and mTRAQ-labeled peptides and for transfer learning can be found in the "iRT_prediction" folder of our GitHub. The files required to regenerate the analyses and model-creation (Files_to_repeat_RT_prediction_and_TransferLearning.zip) are also available for download from our MassIVE repository: MSV000097736

Searching data with JMod:

The workflow is: Thermo .raw → centroided .mzML (MSConvert) → JMod search (command line).

1. Install JMod

JMod uses the uv package manager (pip install uv if you don't have it). From a terminal:

git clone https://github.com/ParallelSquared/jmod.git
cd jmod
uv sync --python 3.11
source .venv/bin/activate

On Windows, activate the environment with .venv\Scripts\activate instead. See the JMod README for more installation options, including a GUI.

2. Convert .raw files to .mzML with MSConvert

JMod reads .mzML files, and the spectra must be centroided. Convert with MSConvert from ProteoWizard.

MSConvert GUI:

1. Add the .raw file(s) and set Output format to mzML.
2. Under Filters, choose Peak Picking, set Algorithm to Vendor and MS Levels to 1-, then click Add. 1- means "MS level 1 and up", so both MS1 and MS2 spectra are centroided.
3. Make sure Peak Picking is the first filter in the list, because vendor peak picking must run on the unmodified vendor data.
4. Click Start.

MSConvert command line (equivalent):

msconvert JD0413.raw --mzML --filter "peakPicking vendor msLevel=1-" -o mzML/

3. Get a spectral library

JMod takes a spectral library as a .tsv in DIA-NN library format (the required columns are shown in JMod's example library, data/filtered_library.tsv). The libraries used in the article are in the MassIVE repository: MSV000097736.

4. Run the search

The example below searches one mTRAQ 3-plexDIA × 3-timePlex run (9 samples in one injection):

python path/to/jmod/run_jmod.py -l path/to/DIANNv1p9_JD0504_506_mTRAQ_HY_jmod.tsv -i path/to/JD0413_re.mzML --iso --num_iso 2 --lib_frac 0.2 --timeplex --num_timeplex 3 --no_ms1_req --atleast_m 2 --plexDIA --tag mTRAQ

You can run it with uv run python ... instead if you haven't activated the environment. To search several files, run the command once per .mzML.

What each option does

Option	What it does	Default	Used in example
-i, --mzml	Centroided .mzML file to search.	required	JD0413_re.mzML
-l, --speclib	Spectral library (.tsv, DIA-NN format).	required	mTRAQ library
--timeplex	timePlex mode. Treats the run as several staggered, overlapping LC separations. Each time channel gets its own RT alignment and search window, and identifications are reported per time channel (time_channel column).	off	on
--num_timeplex	Number of time channels (staggered sample loads) in the run. Must match how the data was acquired.	0	3
--plexDIA	plexDIA mode. Precursor q-values are combined across the label channels of a plexDIA set (BestChannel_Qvalue), so a precursor confidently identified in one channel is also reported in its sibling channels. Use together with --tag.	off	on
--tag	Mass tag used to label the samples. JMod expands every library precursor into one entry per channel of that tag. The name must match a tag defined in JMod's src/MassTags/, e.g. mTRAQ (Δ0/Δ4/Δ8), PSMtag_9plex, diethyl_3plex. Leave unset for label-free data.	None (label-free)	mTRAQ
--iso	Model MS2 fragment isotopes. Each library fragment is expanded into its isotopic peaks, and those peaks are fit jointly with the observed spectrum.	off	on
--num_iso	Number of isotopic peaks per fragment when --iso is on (2 = monoisotopic + M+1).	2	2
--lib_frac	Candidate pre-filter. A precursor is only fit in a scan if its matched fragments account for more than this fraction of its total library fragment intensity. Lower values let more candidates through, which is more sensitive but slower.	0.5	0.2
-m, --atleast_m	Candidate pre-filter. A precursor must match more than this many of its top-10 library fragments, so --atleast_m 2 requires at least 3 matches.	3	2
--no_ms1_req	Don't require a matching MS1 precursor peak for a candidate to be considered. By default, an MS1 peak is required.	MS1 required	MS1 not required

Other useful options

Option	What it does	Default
-o, --output_folder	Folder for the results. If omitted, a <mzML name>_results folder is created next to the .mzML. If that folder already exists, a timestamp is appended so earlier results are never overwritten.	next to the .mzML
-p, --ppm	MS2 fragment m/z tolerance (ppm).	10
--ms1_ppm	MS1 m/z tolerance (ppm). 0 means JMod estimates it from the data.	0 (estimated)
--user_rt_tol with --rt_tol	Force a fixed RT tolerance (in minutes, set by --rt_tol) instead of the one JMod learns from the data.	off / 0.5
-t, --threads	Number of CPU threads.	10
--config_json	Load all parameters from a JSON file. Every search writes its configuration to outputs/config.json, so a search can be repeated exactly with --config_json path/to/outputs/config.json.	—

Adapting the example to other experiment types in this article

Experiment	Change to the example command
Label-free DIA (no multiplexing)	Remove --timeplex --num_timeplex 3 --plexDIA --tag mTRAQ and use an LF library.
Label-free timePlex	Remove --plexDIA --tag mTRAQ and use an LF library.
plexDIA only	Remove --timeplex --num_timeplex 3.
Other plexDIA tags (e.g. 9-plex)	Set --tag to the matching tag in src/MassTags/ and use the matching library.

Boolean options (--iso, --timeplex, --plexDIA, --no_ms1_req, …) are switches and take no value. To confirm a run used the parameters you intended, check the Namespace(...) line near the top of Log.log, which records the full configuration actually used.

5. Outputs

Results are written to the <mzML name>_results folder:

• filtered_IDs.csv: target precursors passing 1% FDR (BestChannel_Qvalue < 0.01), with all columns.
• filtered_IDs_parquet_columns.parquet: the same IDs with a smaller set of key columns (sequence, charge, channel, time_channel, q-values, protein, and quantities).
• outputs/all_IDs.csv: all scored targets and decoys before FDR filtering.
• Summary.txt: precursor and protein ID counts per channel.
• Log.log: full log of the search, including any warnings or errors.
• outputs/config.json: the parameters used, reusable with --config_json.

A full description of the output columns is in the JMod documentation.