By Caimei Li, Senior Scientist at Creative Proteomics (LinkedIn). RUO. Last updated: January 2026
Introduction
Biomarker discovery in metabolomics lives at the intersection of speed, coverage, and quality. Early exploration teams need broad detection across chemical classes to generate hypotheses quickly, yet they also need pathways to convert promising signals into robust assays that stand up to scrutiny. This guide explains how targeted and untargeted approaches work together to serve those goals under a quality-aware R&D context, and it previews a practical decision framework mapping study stage, constraints, and QC readiness to recommended strategies.
We'll define the roles of targeted vs untargeted metabolomics and the hybrid pathways that link them. We'll then walk through decision criteria (study stage, constraints), QA/QC practices (pooled QC, SST, batch correction), and validation checkpoints aligned with ICH M10 principles for bioanalytical methods. The emphasis is on discovery efficiency/coverage and practical risk mitigation so you can move from untargeted "feature soup" to targeted, auditable biomarker quantitation.
- This is a biomarker decision guide—not a general targeted-versus-untargeted overview.
- We compare both approaches across performance, identification confidence (MSI), QC gates, and audit-ready deliverables so teams can choose fit-for-purpose strategies at each study stage.
Definitions and Use Cases
Targeted metabolomics overview
Targeted metabolomics quantifies a predefined panel of metabolites using LC–MS/MS (MRM/PRM), GC–MS, or NMR with authentic standards. It delivers high-confidence results (often MSI level 1 with matched retention time and MS/MS spectrum) and enables absolute or calibrated relative quantification. Typical applications include verifying candidates from discovery studies, validating biomarkers for translational research, and generating data suitable for regulated contexts. For platform specifics, see the targeted service overview in Creative Proteomics' targeted metabolomics page.
Untargeted metabolomics overview
Untargeted metabolomics uses high-resolution MS (Orbitrap or Q-TOF) and GC–MS to survey thousands of features across diverse chemical classes. The goal is discovery—maximizing coverage to find unexpected pathway signals or novel candidates. Identification confidence spans MSI levels 2–3 for most features, with 10–20% typically annotated depending on matrix and acquisition. For practical workflow components, refer to the untargeted metabolomics service overview and the broader metabolomics services.
For a general, vendor-neutral primer on the two approaches, see our overview of targeted vs untargeted metabolomics approaches.
When each is appropriate
Use untargeted when your priority is discovery breadth—early-phase studies, limited prior hypotheses, or when you need to explore pathway-wide changes. Use targeted when you have a defined candidate list that requires quantitation, precision/accuracy testing, and study-to-study comparability. Many teams run hybrid strategies: untargeted HRMS to generate candidates and prioritize signals, then targeted LC–MS/MS to verify and ultimately validate biomarkers for decision-making and potential qualification.
To plan a fit-for-purpose study design and QC package, see our metabolomics services.
Analytical Performance and Coverage in targeted vs untargeted metabolomics
Precision, accuracy, LOD/LOQ
In targeted LC–MS/MS panels, quantitative precision typically falls within ≤20% CV and accuracy within 80–120%, with LOD/LOQ ranges spanning nM to µM for many analytes. Recent multi-analyte work reported LODs from ~1.4 nM to 10 mM across hundreds of metabolites and linear ranges with R² > 0.99, consistent with bioanalytical expectations. See benchmarks summarized in literature such as Zhang et al. (2024) on LC–MS/MS multi-metabolite assays and context from Sands et al. (2021) on LC–MS QC factors.
Untargeted studies aim for stable QC behavior rather than strict panel-level accuracy; post-correction QC CVs in the 20–30% band are commonly accepted in discovery contexts provided biological separation is preserved. Annotation confidence is addressed separately via MSI.
Table — Platform coverage and performance benchmarks
| Platform | Typical coverage | Sensitivity (LOD/LOQ) | Identification confidence (MSI) | Notes |
|---|---|---|---|---|
| LC–MS/MS (MRM/PRM) | Defined panels (50–1000 analytes) | nM–µM (e.g., 1.4 nM–10 mM across large panels) | Level 1 feasible with standards | ICH M10-aligned validation for biomarkers |
| HRMS LC–MS (Orbitrap/Q-TOF) Untargeted | Thousands of features | Matrix-dependent; feature-level sensitivity varies | Typically Level 2–3; 10–20% annotated | Add MS/MS and CCS to raise confidence |
| GC–MS/GC×GC–MS | Volatile/semi-volatile classes; more peaks with GC×GC | ng-level detection typical | MSI Level 2 supported by EI libraries | Orthogonal to LC–MS |
| NMR | ~50–200 metabolites | ≥µM sensitivity | Structural specificity; MSI less applicable | High reproducibility, lower sensitivity |
Platform coverage differences
LC–MS offers the broadest compromise of coverage and throughput across polar and many non-polar metabolites; GC–MS excels for volatile/semi-volatile compounds and can triple peak counts with GC×GC versus 1D GC–MS; NMR provides unmatched structural specificity but lower sensitivity, typically resolving tens to hundreds of metabolites in common biofluids. For practical considerations on instrument setup and class coverage, vendor-neutral overviews include Thermo Fisher's metabolomics workflows.
Identification confidence (MSI)
The MSI framework sets levels from 1 (confirmed with authentic standard) to 4 (unknown). In untargeted datasets, annotation rates of 10–20% at MSI 1–2 are typical, improving with rich MS/MS, retention index/RT prediction, and ion mobility CCS. For background on MSI practice and community standardization, see the NIEHS consortium untargeted metabolomics overview and reports on confidence improvements with CCS such as Analytical Chemistry (2023) on CCS-informed identification.
| MSI level | What it supports in biomarker discovery | Required evidence (what you must provide) | Next step (how to escalate evidence) |
|---|---|---|---|
| MSI 1 — Confirmed ID | Publication- and decision-ready identification for named biomarkers; suitable for verification and validation claims (fit-for-purpose). Strongest basis for cross-study comparison and targeted assay development. |
|
|
| MSI 2 — Putatively annotated | Strong candidate prioritisation for biomarker discovery; acceptable for hypothesis generation and pathway-level interpretation when caveats are stated. Often sufficient to enter a verification shortlist. |
|
|
| MSI 3 — Putatively characterised compound class | Useful for pathway or chemical-class signals (e.g., "a phosphatidylcholine species changes"), early discovery triage, and exploratory biology. Not suitable for naming a specific biomarker without follow-up. |
|
|
| MSI 4 — Unknown feature | Supports only feature-level discovery signals (statistical separation, clustering, drift checks). Good for "something changes" but insufficient for mechanistic claims or biomarker naming. |
|
|
QA/QC and Regulatory Readiness
Pooled QC, SST, randomization
Pooled QC injections are the backbone of discovery-stage quality control. Surveys of LC–MS untargeted studies show pooled QC frequencies commonly every 6–10 samples, with ≥10% QC injections recommended for small runs to condition the system and support QC-based correction. See Broeckling et al. (2023) survey findings and recommendations summarized in QComics (2024). SST should confirm mass accuracy (≈2–5 ppm), RT stability, sensitivity via test mixtures/internal standards, and acceptable peak shapes before the batch proceeds; practical setup guidance is outlined in the LC–MS configuration resource.
Randomize samples across the run, interleave QCs, and include blanks to monitor carryover. Document SST pass/fail and QC schedules for auditability.
Batch correction and acceptance criteria
QC-based correction methods such as QC-RLSC/QCRSC, SERRF, and SVR reduce drift; ComBat (location-scale) adjusts for batch effects; newer approaches like CordBat further improve correction performance. Acceptance checks include tight QC clustering (PCA/t-SNE), QC CV/RSD reduced into the ≤20–30% band, minimized batch separation, and preserved biological differences. See method resources like qcrlscR and evaluations such as CordBat in Analytical Chemistry (2022).
Table — QA/QC acceptance criteria (discovery context)
| Checkpoint | Typical criteria (discovery context) | Rationale | Tools | Minimum reporting evidence (what to show) |
|---|---|---|---|---|
| Pooled QC cadence | Every 6–10 injections; ≥10% QCs for small runs | Drift tracking; conditioning; correction readiness | Study design; LIMS; QC-MXP |
|
| QC reproducibility | QC CV/RSD ≤20–30% post-correction; tight QC clustering (PCA) | Validates stability and correction effectiveness | TIGER; QC-MXP; MetaboAnalyst |
|
| SST thresholds | Mass accuracy ≈2–5 ppm; RT tolerance (method-dependent); carryover checks | Ensures instrument readiness and data quality | Vendor tools; LC–MS setup guidance |
|
| Batch correction performance | Reduced batch separation; low MAPE on QCs; biological separation preserved | Confirms harmonisation without overfitting | qcrlscR; SERRF; ComBat; CordBat |
|
Typical ranges are illustrative and should be tuned to matrix, platform, and study risk; always report the evidence pack above so results are reviewable.
ICH M10 and biomarker qualification
ICH M10 governs bioanalytical method validation and study sample analysis for chromatographic assays—relevant when you convert discovery candidates into targeted biomarker quantitation. For endogenous analytes, M10 outlines surrogate matrix/analyte strategies, standard addition, and background subtraction, alongside requirements for selectivity, calibration, accuracy, precision, matrix effects, stability, and incurred sample reanalysis. See the official guidance from EMA (Step 5 PDF, 2022) and the FDA guidance page.
Data Workflows and Tooling
Targeted quantification workflow
A practical targeted pipeline: develop transitions (MRM/PRM) with isotopic standards; optimize chromatography; acquire LC–MS/MS data with calibration curve fitting; run QC ladders (LLOQ, low/mid/high) across batches; evaluate accuracy, precision, matrix effects, and stability; and produce audit-ready reports. Skyline's audit logging and Panorama integration help lock methods and preserve traceability.
Untargeted discovery pipeline
A typical untargeted workflow includes study design and randomization; SST; pooled QC scheduling; HRMS/GC–MS acquisition (DDA/DIA/AIF); preprocessing (peak picking, alignment, deconvolution, blank masking) in XCMS/MZmine/MS-DIAL; QC-based normalization and batch correction; annotation using MS/MS libraries and, where available, CCS; and downstream statistics/pathway analysis in MetaboAnalyst. For deeper guidance, see Chen et al. (2022) on metabolomics analysis workflows and resources on untargeted preprocessing strategies and data normalization methods.
Software updates and audit-ready outputs
Audit-ready outputs depend on documenting software versions, parameter sets, and changes over time. Skyline provides an interactive audit log; XCMS/MZmine/MS-DIAL support comprehensive parameter export and project files; nPYc-Toolbox enables standardized QC workflows; MetaboAnalyst 6.0 supports reproducible downstream analysis. Capture raw vendor files plus mzML, feature tables (CSV/mzTab), MS/MS library versions, and a batch correction report.
Table — Audit-ready deliverables checklist
| Deliverable | Description | Tool(s) | Notes |
|---|---|---|---|
| Raw data + open formats | Vendor files and mzML exports | Instrument vendor; ProteoWizard | Preserve originals and conversions |
| Processed feature tables | Peak-picked/aligned features (CSV/mzTab) | XCMS/MZmine/MS-DIAL | Include parameter files and versions |
| QC and SST logs | Instrument settings, calibration, pass/fail, QC schedule | nPYc-Toolbox; lab LIMS | Enables traceability and corrections |
| Batch correction report | Method (QC-RLSC/SERRF/ComBat/CordBat), parameters | R/Python scripts; qcrlscR | Show pre/post QC CVs and PCA clustering |
| Targeted method + audit log | Skyline document with audit trail; calibration curves | Skyline; Panorama | Lock method version and transitions |
| Analysis summary | PDF/HTML with acceptance criteria and outcomes | MetaboAnalyst; R Markdown | Executive-ready overview |
Decision Framework: When to Use Each
Table：Biomarker decision matrix
| Study stage | Best-fit approach | Primary success metric | Minimum QC gates (fit-for-purpose) | ID evidence target (MSI) | Deliverables required (audit-ready) |
|---|---|---|---|---|---|
| Discovery (Hypothesis generation) | Untargeted (HRMS LC–MS ± GC–MS) | Coverage and biological signal detection (pathway-level separation) |
| MSI 2–3 (with a plan to escalate candidates) |
|
| Verification (Confirm candidates) | Hybrid: Untargeted + targeted confirmation (PRM/MRM) for a subset | Confirmation and prioritisation (reduce false positives; confirm chemistry) |
| MSI 1–2 (move top hits toward Level 1 where possible) |
|
| Validation (Quantitation for decisions) | Targeted (LC–MS/MS MRM/PRM, GC–MS, or NMR with standards) | Quantitation readiness (precision, accuracy, cross-batch comparability) |
| MSI 1 (authentic standard confirmation for reported biomarkers) |
|
Criteria by study stage and constraints
Worked example
A compact, reproducible pipeline: 120 human plasma samples (60 v. 60), pooled QC every 6 injections (≈10% of injections), SST: mass accuracy ≤3 ppm and RT drift ≤0.2 min. Preprocess with XCMS (parameterized script), apply QC-based SERRF then ComBat for cross-batch harmonization; post-correction pooled-QC CV median ≈18% and PCA shows QCs tightly clustered with biological groups preserved. From untargeted candidates (top 50 by adj. p-value), 18 were progressed to scheduled MRM; pilot MRM showed 85% passing CV ≤20% and method lock. Example data and scripts: public dataset MTBLS235 on MetaboLights and XCMS/SERRF example scripts jorainer/xcms-preprocessing (GitHub) and jorainer/xcms-gnps-large-scale (GitHub).
- Discovery-first with limited hypotheses and a need for coverage: favor untargeted HRMS/GC–MS with rigorous QC and planned batch correction.
- Verification/validation with defined candidates and quantitative goals: use targeted LC–MS/MS (MRM/PRM), GC–MS, or NMR with standards and ICH M10-aligned checks.
- Sample size and matrix complexity: if batch sizes are large or matrices are challenging, consider hybrid designs that confirm a subset in PRM/MRM early.
Cost and timeline trade-offs
Untargeted runs can scale quickly but require time for preprocessing, correction, and annotation; targeted methods demand upfront development (transition selection, CE optimization) but deliver faster, audit-ready quantitation once locked. Scheduling MRM and focusing on class-specific panels can reduce runtime and costs.
Table — Common failure modes in biomarker discovery metabolomics (symptoms → causes → fixes)
| Symptom (what you see) | Likely cause (why it happens) | Fix / prevention (what to do next) |
|---|---|---|
| QC samples drift over time (signal/RT shifts; QC points "walk") | Source contamination, column ageing, temperature instability, ion source fouling; insufficient QC cadence | Tighten SST; clean source/replace consumables; stabilise temperature; increase/regularise pooled QC placement; apply QC-based correction (report pre/post) |
| QC CV/RSD remains high after correction | Poor peak integration/alignment; unstable ionisation; inappropriate correction model; heterogeneous pooled QC | Re-check peak picking/alignment parameters; exclude unstable features; re-make pooled QC (matrix-matched); switch/compare correction methods (QC-RLSC/SERRF/SVR); confirm correction isn't overfitting |
| Strong separation by batch/plate (batch dominates PCA) | Run-order confounding; different operators/reagents; instrument state changes; inadequate randomisation | Enforce randomisation; block design by group; add bridge QCs across batches; standardise reagents; document instrument state; use batch correction (ComBat/CordBat) and show "biology preserved" |
| Many features also present in blanks | Solvent contamination; plasticisers; carryover; background ions | Implement blank masking rules; use process blanks; change solvents/lines; increase wash steps; identify recurring contaminants and blacklist them |
| Carryover observed (blank-after-high shows target peaks) | Autosampler needle/seat contamination; column memory; insufficient wash; sticky analytes | Add/extend needle wash; use strong wash solvents; insert blank-after-high; service needle seat; adjust gradient; consider column swap for high-memory chemistries |
| Unexpected loss of polar metabolites (weak early-eluters) | Sample prep losses; poor retention/ionisation; ion suppression; wrong column chemistry | Review extraction solvent ratios; add isotopically labelled IS for polar class; consider HILIC; optimise LC conditions; reduce salt; evaluate matrix effects (post-column infusion/standard addition) |
| Lipid features unstable or poorly annotated | In-source fragmentation; isomer complexity; insufficient MS/MS depth; wrong adduct handling | Use lipid-class internal standards; acquire class-appropriate MS/MS; enforce adduct rules; consider ion mobility/CCS support; confirm key lipids via targeted PRM/MRM |
| Annotation inconsistent across runs (same feature gets different IDs) | Library/version differences; parameter drift; m/z tolerance mismatches; inadequate MS/MS evidence | Lock library/version; record software + parameters; harmonise tolerances; prioritise MSI evidence packs; confirm top hits with standards/targeted MS/MS |
| Significant group differences disappear after QC/batch correction | Over-correction removing true biology; confounded design; small sample size | Validate design/randomisation; compare multiple correction methods; check "biology preserved" plots; use sensitivity analyses; verify candidates in an independent subset with targeted confirmation |
| Top "biomarker" is likely a mis-ID / isomer | Isobaric/isomeric interference; poor chromatography; MSI 2–3 treated as MSI 1 | Treat MSI 2–3 as candidates only; acquire confirmatory MS/MS; improve chromatographic separation; purchase standards; move to targeted assay for final claims |
| Matrix effects / ion suppression suspected (signal depends on matrix/run position) | Co-eluting phospholipids/salts; variable sample quality (hemolysis/lipemia); insufficient clean-up | Use isotope-labelled IS; evaluate with post-extraction spike / standard addition; optimise chromatography; add/adjust clean-up; stratify/flag problematic samples; document interference policy |
| Results hard to reproduce / not reviewable | Missing run-order/QC logs; undocumented parameter changes; no audit trail | Provide evidence pack: run order map, SST/QC logs, parameter files, software versions, pre/post correction plots; lock targeted methods (Skyline audit log) |
Hybrid strategies and risk mitigation
Hybrid pipelines connect discovery to validation: PRM confirmations of untargeted hits, SQUAD-style single-injection MS1 discovery paired with targeted MS2, and cross-batch harmonization using pooled QC-driven corrections. These strategies mitigate the risk of false positives and ensure consistency when studies scale or repeat.
Hybrid Pipeline: Discovery to Validation
Untargeted → targeted assay development
Start with candidate triage informed by statistics and biological plausibility, acquire confirmatory MS/MS (DDA/DIA/PRM), use RT/CCS predictions, and select robust precursor/product transitions. Build a scheduled MRM method, run a pilot with a QC ladder (LLOQ/low/mid/high), and iterate until precision/accuracy targets and matrix effects are acceptable. Document parameters and lock the method version.
Validation and cross-batch harmonization
Apply ICH M10 elements (selectivity, calibration, accuracy, precision, matrix effects, stability, and incurred sample reanalysis), then harmonize across batches via QC-based LOESS plus ComBat or CordBat. Acceptance criteria typically include QC CV ≤20–30%, minimal batch separation in PCA, and preserved biological signals.
Outsourcing and vendor selection
Disclosure: Creative Proteomics is our product. Example-only, not a performance claim.
Vendor questions to ask (biomarker discovery metabolomics)
- QC design: What is your pooled QC strategy (source, aliquoting, and injection cadence), and how do you report drift?
- SST gates: What system suitability checks do you run before each batch, and what triggers a stop/restart?
- Blanks & carryover: Which blanks are included (solvent, process), and how do you monitor and document carryover?
- Matrix effects: How do you assess ion suppression/enhancement (e.g., post-extraction spike, standard addition, infusion tests), and when do you recommend isotope-labelled standards?
- Annotation confidence: How do you assign MSI confidence levels, and what evidence is provided for MSI 1 vs MSI 2–3 calls?
- MS/MS depth: When and how do you acquire confirmatory MS/MS (DDA/DIA/PRM), and what is your plan for isomers/interferences?
- Batch correction: Which normalisation/correction methods do you use (e.g., QC-LOESS/SERRF/ComBat), and do you provide pre/post plots plus parameter files?
- Deliverables: Will you provide raw vendor files plus open formats (mzML), feature tables with a field dictionary, and an audit-ready QC/SST pack?
- Reanalysis pathway: Can you support reanalysis and the bridge to targeted assays (PRM/MRM), including method version lock and an "evidence pack" for shortlisted biomarkers?
- Change control & retention: How are software versions, library versions, and processing parameters tracked, and what are your data retention and access-control terms?
If you want, we can review your study stage and constraints and return a fit-for-purpose plan covering QC gates, evidence level (MSI), and deliverables.
For a hybrid pipeline, selecting a vendor with LC–MS/MS MRM/PRM, HRMS Orbitrap/Q-TOF, GC–MS, and NMR plus bioinformatics support simplifies coordination. As an example, Creative Proteomics lists targeted and untargeted metabolomics platforms and resources, including triple quadrupole/QTRAP for MRM, Orbitrap/Q Exactive for HRMS discovery, and workflow pages on untargeted preprocessing and normalization methods. In outsourcing, specify deliverables contractually (raw and open-format data, QC/SST logs, batch correction report, Skyline audit log, method/version lock) and benchmark QC expectations (e.g., pooled QC cadence every 6–10 injections; post-correction QC CV ≤20–30%) using community standards.
Conclusion
Targeted vs untargeted metabolomics is not an either–or decision; it's a staged strategy. Use untargeted to maximize coverage and find signals quickly, and targeted to quantify, verify, and validate biomarkers with audit-ready rigor. Across both, the foundation is QA/QC: pooled QC cadence, SST checks, robust batch correction, and transparent reporting of MSI confidence. When you bridge discovery to validation, apply ICH M10-aligned checks and lock methods and versions to ensure reproducibility and cross-batch consistency. That's the path to discovery velocity without sacrificing reliability.
If you're moving from discovery signals to verification or targeted validation, our metabolomics services team can help scope the right workflow, QC gates, and audit-ready deliverables.
