The problem the developers set out to solve is one that any practicing metabolomics researcher will recognise. Instruments from SCIEX, Agilent, Waters, Thermo, and Bruker each produce data in proprietary formats that lock laboratories into specific software ecosystems, complicating cross-laboratory harmonisation and restricting long-term accessibility and reuse under FAIR data principles. Converting files to the community-standard open format mzML is a necessary first step, but it is only the beginning. Metabolic phenotyping experiments are also vulnerable to systematic variation arising from sample preparation, instrument performance drift, and batch effects, and even the choice of processing tool and its parameterisation introduces additional variance. Robust, standardised quality-control procedures are therefore essential, yet until now no single open tool combined vendor file conversion, automated peak integration, full auditability, and a documented quality-control layer in one scriptable environment.

MStargetR, developed by Harrison Szemray, Luke Whiley, and colleagues at Murdoch University and collaborating institutions in Australia and Colombia, takes a pragmatic architectural approach: rather than reinventing established tools, it builds on them. The package is organised into three core modules that map directly onto the sequential stages of a targeted multiple reaction monitoring (MRM) or selected reaction monitoring (SRM) experiment. The first, msConvertR, interfaces with ProteoWizard’s msConvert tool to convert proprietary vendor files into mzML format, accepting raw data from SCIEX, Thermo, Agilent, Bruker, and Waters instruments as well as files already in open formats. The second, PeakForgeR, performs retention time alignment, peak boundary optimisation, and automated integration. The third, qcCheckR, handles missing value imputation, internal standard normalisation, concentration calculation, signal drift and batch correction, filtering, and reporting.

The technical heart of the pipeline lies in how PeakForgeR handles peak integration. For each analytical batch, a lightweight algorithm identifies the peak apex and boundaries for each transition window, using the median apex retention time localisation across quality-control samples to define a plate-wide reference. These pre-defined boundaries are then pushed to the Skyline command-line interface for peak area integration, ensuring that consistent integration windows are applied across all samples rather than executing independent per-sample peak detection that could introduce variable integrals. Crucially, this approach produces a fully populated .sky document for every analytical plate. That document preserves every chromatogram, retention time annotation, and peak boundary used to generate the result table, and it can be reopened directly in the free Skyline graphical interface, where analysts can visually inspect individual peaks, manually adjust integration boundaries where automated detection falls short, and re-export corrected results that qcCheckR reimports without rerunning any upstream steps.

This handoff between automation and human oversight is arguably the package’s most distinctive design choice. Automated peak picking becomes a transparent first pass that the analyst remains in control of, while the .sky document itself becomes a shareable, auditable record of every integration decision behind a reported concentration matrix. It is a deliberate attempt to bridge the gap between proprietary vendor platforms, which offer rich visual inspection but are platform-locked and licence-dependent, and open-source R packages, which offer scriptability but have traditionally left the audit trail behind. The developers also provide a standalone batch correction module, batchCorrectR, that exposes the signal drift and batch correction routines for use on any tabular metabolomics dataset, including data processed through other software entirely.

On the quality-control front, qcCheckR offers three well-established correction methods that laboratories can select according to their own practices: quality-control-based random forest signal correction (QC-RFSC), which captures non-linear drift that linear models cannot; quality-control-based robust LOESS signal correction (QC-RLSC); and ComBat, an empirical Bayes method suitable for datasets that lack QC samples altogether. Missing values below a user-specified instrumental limit of detection are handled by imputing half the minimum observed value for each feature, a strategy designed to maintain data integrity. The module generates comprehensive outputs in both HTML and Excel formats, with the HTML report including summary tables of key quality metrics and interactive visualisations such as principal components analysis plots and control charts that allow dynamic exploration of data quality at the level of individual metabolites.

Reproducibility is engineered into the package at the infrastructure level. All third-party tools, including msConvert and Skyline, are bundled within version-pinned Docker images, so that the exact software environment travels with the analysis and eliminates complex dependency management. For high-performance computing environments that prohibit Docker, MStargetR can be configured to use Apptainer instead. The pipeline runs natively on AMD64 architecture across Windows and Linux, with macOS and ARM64 users currently needing a virtual machine, a limitation the team is actively working to address. For analysts who prefer not to write code at all, a Shiny-based graphical user interface provides code-free access to the full pipeline, and a Windows installer delivers the application directly from the Start menu, requiring only a container runtime alongside it.

The performance evaluation is striking. Applied to a published targeted lipidomics dataset of 128 human plasma samples targeting 1,161 lipid species with 73 stable isotope-labelled internal standards, MStargetR processed all samples end-to-end, recovering all 1,161 targeted lipid features. Of these, 949 features (81.7 percent) were detected and returned a relative standard deviation below 30 percent across replicated long-term reference quality-control samples, and 823 features met the more stringent RSD threshold of 10 percent, a figure the authors emphasise reflects combined effects of biological absence, method sensitivity, and deliberately stringent quality criteria rather than detection failure. Because per-transition precision metrics and the corresponding chromatograms are both retained, the reason any individual target falls below threshold can be inspected directly rather than inferred.

Scalability testing on a standard workstation showed that total runtime grew linearly from 14.9 minutes at 84 samples to 302.5 minutes at 4,200 samples, with per-sample processing time falling to just 4.1 seconds at larger batch sizes as fixed overhead was distributed across more injections. CPU utilisation plateaued around 62 to 67 percent once enough plates were available to keep the parallel workers supplied, and memory consumption was dominated by per-plate parallelism overhead rather than per-sample data accumulation. Predictable, linear scaling means laboratories can reliably estimate processing time for large cohort studies from the reported per-sample rate, a practical consideration as metabolomics studies grow to thousands of specimens.

The developers are careful to position MStargetR’s contribution as integration rather than algorithmic superiority. Quantification currently relies on single-point internal standard calculation rather than multi-point calibration curves, and dedicated MRM processors remain more complete in that respect, while the browser-based MRMPro remains stronger for rapid multi-user manual curation of very large cohorts. But no existing open tool, the authors argue, combines vendor conversion, automated integration, a re-importable audit artefact, and a documented quality-control and batch correction layer within one version-pinned, R-native environment. Future work will expand supported input formats, add multi-point calibration for absolute quantification across wider dynamic ranges, and tighten integration with public repository deposition. For a field whose credibility increasingly depends on whether results can be independently reproduced, MStargetR represents a meaningful step toward making rigorous, auditable targeted metabolomics the default rather than the exception.

Subject of Research: A reproducible, containerised R workflow for end-to-end targeted MRM/SRM mass spectrometry data processing

Article Title: MStargetR: a reproducible, containerised workflow for end-to-end targeted (MRM/SRM) mass spectrometry data processing in R

Article References: Szemray, H., Nambiar, V., Hicks, D., Lodge, S., Wist, J., Lawler, N. G., & Whiley, L. (2026). MStargetR: a reproducible, containerised workflow for end-to-end targeted (MRM/SRM) mass spectrometry data processing in R. Metabolomics, 22(5), Article 160. https://doi.org/10.1007/s11306-026-02541-2

Image Credits: AI Generated

DOI: 10.1007/s11306-026-02541-2

Keywords: metabolomics, mass spectrometry, MRM, SRM, lipidomics, reproducibility, R software, quality control, batch correction, FAIR data, open-source software, Docker