MsTargetPeaker: a quality-aware deep reinforcement learning approach for peak identification in targeted proteomics

MsTargetPeaker incorporates a deep reinforcement learning agent and Monte Carlo tree search to locate target peak regions in targeted mass spectrometry. The agent was trained with proximal policy optimization on a big collection of targeted MS datasets containing around 1.7M peak groups. During the training, we established a gymnasium environment for the agent to move peak boundaries to locate target signals. To define optimal peaks, we designed a reward function incorporating our previously developed TMSQE quality scoring. Thus, the agent can learn autonomously to find high-scoring peak regions without using maunally annotated peaks. In the end, the training process took about 200M timesteps to reach performance plateau.

The peak search procedure in MsTargetPeaker was performed using Monte Carlo tree search guided by this agent to enhance the generability, especially for unseen datasets. To further enhance the precision on ambiguous peaks, additional search rounds were appended trying to locate peak regions enclosing true target signals.

After running the peak search, the generated peak csv file can be imported into Skyline for manual re-evaluation or peak integration. MsTargetPeaker also provides a peak reporter to generate interpretable peak quality reports.

Currently, MsTargetPeaker supports peptide MRM/PRM data.

Installation

MsTargetPeaker was built as a Python package. You can use the following command to install the package.

pip install MsTargetPeaker

After you install mstargetpeaker, you can use mstarget-peaker and mstarget-reporter as the command line tools for identification of peak regions and assessment of the peak quality. Use --help or -h to see detailed argument descriptions.

Input Data Format

MsTargetPeaker currently accepts chromatogram data in tab-separated value (TSV) format. This chromatogram file can be exported via Skyline.

The required nine column headers for chromatograms are listed as follows.

FileName	PeptideModifiedSequence	PrecursorCharge	ProductMz	FragmentIon	ProductCharge	IsotopeLabelType	Times	Intensities

Usage

Use the following command to run MsTargetPeaker to search peak regions in chromatograms.

MsTargetPeaker <chromatogram_tsv> <output_peak_boundary_csv>

With this command, MsTargetPeaker takes the first chromatogram TSV file as input and outputs the resulting peak regions to the CSV file specified in the second argument. The resulting peak CSV file can be imported into Skyline to update peak regions in the chromatograms.

The full arguments are shown below:

MsTargetPeaker [-h] [--speed SPEED] [--search SEARCH] [--config CONFIG] [--picked PICKED] [--process_num PROCESS_NUM] [--internal_standard_type {heavy,light}] [--device DEVICE] chromatogram_tsv output_peak_boundary_cs

Argument		Description	Value Type	Default Values
INPUT
chromatogram_tsv		The chromatogram TSV file path	File path	no default
OUTPUT
output_peak_boundary_csv		The output peak boundary CSV file path	File path	no default
Options
--help	-h	Show the detailed argument list	(no value)	unset
--version	-v	Display the package version	(no value)	unset
--speed	-s	The speed mode of UltraFast (10X), Faster (5X), Fast (2X), or Standard (1X speed). This can be customized in the config file.	string	UltraFast
--mode	-m	The search mode using the parameter set of MRM or PRM. This can be customized in the config file.	string	MRM
--prescreen	-pre	Prescreen peak regions for better peak boundaries as initial state.	int	50
--internal_standard_type	-r	Set the internal standard reference to heavy or light ions.	{heavy, light}	heavy
GROUPING
--process_num	-p	The parallel process number to search peak regions	integer	4
--device	-d	Use cpu or cuda device for peak picking.	string	auto
Incremental Peak Search
--picked		The previously picked boundaries for incremental peak search.	File path	unset
--start_round	-sr	Specify the starting MCTS round in the config file. This is useful for incremental peak search.	int	1
--end_round	-er	Specify the ending MCTS round in the config file. This is useful for incremental peak search.	int	7

Incremental Peak Search

MsTargetPeaker supports incremental peak search from a previously identified peak boundary csv file (You may use the peak boundary results from Skyline or other peak identification tools). To further reduce the search time, users can initially use --speed=SuperFast to have a quick result. Then, specify --picked={the peak csv file} with the --start_round=4 to start the search with parameters of the 4th to the last round of MCTS. With this setting we can re-search peak groups which rewards failed to pass the threshold set in the config file.

Configuration

The default configuration file is MsTargetPeaker.cfg. You may customize this file to suit your preferences.

Quality Reporter

The reporter can be run independently to generate the following five reports:

1. Transition quality files in a folder.
2. An Excel file containing two sheets: sample quality and replicate group quality.
3. A PDF showing chromatogram plots.
4. A PDF swhoing the probability density functions of peak start and end for each target.

To run the quality reporter, use the following command:

MsTargetReporter [-h] [--internal_standard_type {heavy,light}] [--top_n_fragment TOP_N_FRAGMENT] [--group_csv GROUP_CSV] [--output_chromatogram_pdf] [--chromatogram_dpi CHROMATOGRAM_DPI]
                 [--chromatogram_nrow CHROMATOGRAM_NROW] [--chromatogram_ncol CHROMATOGRAM_NCOL] [--chromatogram_fig_w CHROMATOGRAM_FIG_W] [--chromatogram_fig_h CHROMATOGRAM_FIG_H] [--output_mixed_mol] [--reorder_by_group]
                 chromatogram_tsv peak_boundary_csv output_folder

The full arguments are shown below:

Argument		Description	Value Type	Default Values
INPUT
chromatogram_tsv		The chromatogram TSV file path	File path	no default
peak_boundary_csv		The output peak boundary CSV file path	File path	no default
OUTPUT
output_folder		The output peak boundary CSV file path	File path	no default
Options
--help	-h	Show the detailed argument list	(no value)	unset
--group_csv	-g	The CSV file containing the replicate group information	File path	unset
--top_n_fragment	-n	Automatically select top N transition ions for reporting the quality	integer	5
Options for Generating Chromatogram Plots
--output_chromatogram_pdf	-pdf	Set for generating chromatogram plots in a file named chromatogram_plots.pdf	File path	unset
--output_mixed_mol	-mix	If set, chromatogram plots for each target molecule will be mixed in one PDF page.	(no value)	unset
--reorder_by_group	-r	If set, target molecule will be reordered based on the replicate group. Only works if the --group_csv is provided.	(no value)	unset
--chromatogram_dpi	-dpi	The dpi of chromatogram plots. Only works when --output_chromatogram_pdf is set.	integer	200
--chromatogram_nrow	-nrow	#Rows per PDF page.	integer	6
--chromatogram_ncol	-ncol	#Columns per PDF page.	integer	6
--chromatogram_fig_w	-figw	The width of a PDF page (in inches)	float	ncol*7
--chromatogram_fig_h	-figh	The height of a PDF page (in inches)	float	nrow*5

Utility Functions

Chromatogram Checking

We noticed that certain exported chromatogram TSV files from Skyline may have unpaired arrays of Time and Intensity between light and heavy ions. Also, as we currently rely on the modified peptide sequence and sample file name to recognize each peak group, it may cause issues if the chromatogram data contain duplicate peptide-sample names.

We provided MsTargetChromChecker to solve these two issues. For the unaligned data points in light and heavy ions, we apply interpolations to make the same number of retention time and its intensity for both light and heavy ions. For duplicated names for peak groups, MsTargetChromChecker appends a suffix to the sample file names. The suffix has a pattern of ::n, where n is a number indicating the duplication number.

Use the following command to run MsTargetChromChecker,

MsTargetChromChecker [-h] chromatogram_tsv output_chrom_tsv

Parallelism

As it takes time to run MsTargetPeaker, we provide MsTargetChromSplitter to split the task into smaller ones. Each splitted task can be run parallelly on different processes or machines. The results from these tasks can then be merged with MsTargetPeakMerger.

MsTargetChromSplitter

MsTargetChromSplitter [-h] -n [number of file] chromatogram_tsv output_folder

The default spliting number is the number of target moleculars (without specifying the -n argument).

MsTargetPeakMerger

MsTargetPeakMerger [-h] input_folder output_csv_file

This MsTargetPeakMerger accepts a folder containing multiple peak csv files and output the merged version of those csv files. You can use MsTargetChromSplitter to split the input chromatogram TSV file, run MsTargetPeaker in parallel on each split file to search for peak regions, and merge the resulting peak CSV files in a folder using MsTargetPeakMerger.

Citation

If you use MsTargetPeaker in your research, please cite:

@article{yang2026mstargetpeaker,
  author = {Yang, C. and Hsiao, Y. C. and Lee, C. C. and Chu, L. J. and Yeh, T. S. and Cheng, P. C. and Tang, P. and Yu, J. S.},
  title = {MsTargetPeaker: A Quality-Aware Deep Reinforcement Learning Approach for Peak Identification in Targeted Proteomics},
  journal = {Molecular \& Cellular Proteomics},
  volume = {25},
  pages = {101523},
  year = {2026}
}