bio-metabolomics-targeted-analysis - SKILL.md Agent Skill

name: bio-metabolomics-targeted-analysis description: Designs and validates quantitative targeted metabolomics assays (MRM/SRM on triple-quadrupole, PRM on high-resolution instruments) to report absolute concentrations. Covers the internal-standard strategy (external cal -> global IS -> standard addition -> stable-isotope-labeled IS), weighted calibration judged by back-calculated %RE not R-squared, ion-ratio quantifier/qualifier confirmation, matrix-effect/recovery characterization, and ICH M10 method validation. Use when quantifying a closed panel of known metabolites with units, building or validating an LC-MS/MS assay, choosing an IS or calibration weighting, or judging whether a reported concentration is trustworthy. For untargeted feature detection see metabolomics/xcms-preprocessing; for group statistics see metabolomics/statistical-analysis; for flux/MID/tracing see metabolomics/isotope-tracing. tool_type: mixed primary_tool: skyline

Version Compatibility

Reference examples tested with: R 4.3+, ggplot2 3.5+, Skyline 23.1+, pandas 2.2+, numpy 1.26+

Before using code patterns, verify installed versions match. If versions differ:

R: packageVersion('<pkg>') then ?function_name to verify parameters
Python: pip show <package> then help(module.function) to check signatures
CLI: <tool> --version then <tool> --help to confirm flags

An absolute concentration requires three inputs the code cannot supply: an authentic reference standard (its certificate-of-analysis purity scales every reported number), a stable-isotope-labeled internal standard that co-elutes with the analyte, and a per-analyte validation record. Without these, the workflow below produces relative peak-area ratios dressed as concentrations.

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Targeted Metabolomics Analysis

"Quantify these specific metabolites and give me concentrations with units" -> Fit weighted calibration curves on authentic standards, normalize each analyte to a co-eluting stable-isotope-labeled internal standard, confirm identity by ion ratio, and report concentrations only within the validated range.

CLI: Skyline builds the small-molecule transition list, integrates peaks, fits weighted calibration, and exports via the Document Grid.
R / Python: post-export curve fitting, back-calculated %RE checks, IS normalization, ion-ratio confirmation, validation metrics.

The Single Most Important Modern Insight -- A Concentration Is a Chain of Cancellations, and Matrix Effects Are the Term That Fails to Cancel

Ion suppression is competition for charge and droplet surface in the electrospray source: a co-eluting matrix component (phospholipids late in a reversed-phase gradient, salts at the void) steals ionization from the analyte. Suppression is a property of the co-elution, not of the analyte, so it is retention-time-dependent and lot-dependent. The only mechanism that truly removes it is a stable-isotope-labeled internal standard (SIL-IS) that sits in the identical droplet at the identical instant: the suppression cancels in the analyte/IS area ratio. An IS that elutes even half a minute away samples a different point on the suppression landscape and injects new error rather than removing it. Every other safeguard in this skill -- weighting, ion ratios, validation -- assumes this cancellation is working; the gap between a solvent calibration curve and a matrix-matched curve is a direct readout of how badly the IS is failing.

Targeted vs Untargeted -- Different Experiments, Not Two Settings

Axis	Untargeted (discovery)	Targeted (quantification)
Analyte set	Open -- everything ionizable	Closed -- a panel defined before acquisition
Output	Relative fold-change; often putative IDs	Absolute concentration for confirmed analytes
Instrument	High-res full-scan / DDA (Orbitrap, QTOF)	Triple-quad SRM/MRM, or high-res PRM
Validation	QA/QC framework (Broadhurst, mQACC)	Full bioanalytical validation possible (ICH M10)
Question	"What changed?"	"How much is there?"

Targeted buys sensitivity and absolute quant by spending scope (only what is on the list is seen) and up-front method development. Common pattern: untargeted discovery -> targeted validation of the hits. Feature detection upstream is metabolomics/xcms-preprocessing; this skill begins once the panel and transitions are defined.

Acquisition Mechanics -- MRM/SRM and PRM

A transition is a precursor-m/z -> product-m/z pair plus a tuned collision energy. On a triple quadrupole, Q1 isolates the precursor, the collision cell fragments it, Q3 isolates one product: the double mass filter is the source of MRM sensitivity. SRM monitors one transition; MRM multiplexes many. Each analyte should carry at least two transitions -- a quantifier (most intense/cleanest, used for concentration) and one or more qualifiers (orthogonal confirmation). PRM replaces Q3 with a high-resolution analyzer that records the full product spectrum in parallel, so transitions are chosen post hoc and isobaric interferences are resolved by exact mass; MRM still wins on absolute sensitivity and very large panels. Dwell time is the signal-accumulation time per transition; cycle time must stay short enough for at least 10-15 points across each chromatographic peak (convention). Scheduled MRM monitors each transition only within a retention-time window so a large panel keeps adequate dwell -- but a peak that drifts out of its window vanishes with no error message, the classic scheduled-MRM failure.

Decision Tree -- Quant Goal -> IS + Calibration + Validation Depth

Goal / situation	Internal standard	Calibration	Validation depth	Why
Clinical / regulated / PK number	One SIL-IS per analyte (13C/15N)	Multi-level weighted curve, judged by %RE	Full ICH M10 (accuracy, precision, MF, recovery, carryover, stability, ISR)	A number driving a decision must carry its evidence
Cross-study quantitative claim	SIL-IS per analyte or per RT/chemical cluster	Multi-level weighted	Accuracy/precision + matrix-factor on QCs	Comparability across runs demands characterized bias
Exploratory research, ranking	Few global IS, or per-class	1/x^2 weighted, low-end %RE checked	Broadhurst/mQACC QC discipline (pooled QC, blanks, RSD filtering)	Relative comparison tolerates residual matrix bias
Dirty matrix, isobaric interferences	SIL-IS + high-res	PRM, post-hoc transitions	Selectivity dominated	Exact-mass product resolves co-eluters a unit-resolution Q3 cannot
Large standardized panel (600+)	Kit-supplied class IS	Single/limited-point (vendor)	Vendor + bridging study before pooling sites	Kit buys comparability and throughput, not per-analyte full-validation accuracy
Carbon source / pathway rate	(tracer, not IS)	--	--	Flux question: hand off to metabolomics/isotope-tracing; MID measures rate, not pool size

The IS rule of thumb: ask how far (in retention time and chemistry) each analyte is from its assigned IS -- that distance is the size of the uncorrected matrix error. 13C/15N at non-exchangeable positions are preferred over deuterium: deuterium causes a small reversed-phase retention shift (the deuterium isotope effect) that can chromatographically separate the IS from its analyte so it stops correcting suppression, and labile deuteriums back-exchange to H. If forced to a deuterated IS, verify co-elution by overlaying analyte and IS chromatograms.

Calibration and Weighting

Weighting	When	Effect
Unweighted (OLS)	Narrow range, near-constant variance	High points dominate; low-end bias on heteroscedastic MS data -- usually wrong
1/x	Moderate range (1-2 orders)	Down-weights high concentrations; restores low-end fit
1/x^2	Wide range (3+ orders), the common LC-MS default	Aggressively down-weights the top; can over-weight the low end -- still compare, do not reflex
Quadratic	Genuine, mechanism-explained curvature (detector saturation)	Never to paper over a bad linear fit

MS detector response is heteroscedastic -- absolute variance grows with concentration -- so weighting models the variance structure (1/x and 1/x^2 are parametric stand-ins for 1/variance). Select empirically: fit candidate weightings, then pick the one minimizing the sum of absolute back-calculated relative error (%RE) across levels, especially the bottom two or three. R-squared is the wrong instrument: it is dominated by high-leverage top points, so a curve with R-squared 0.999 can be +40% biased at the LLOQ. Use a fitted (non-zero) intercept; forcing the line through the origin re-introduces low-end bias. The blank (matrix only) and zero (matrix + IS) are diagnostic, not calibration points.

Build a Weighted Calibration Curve With a Back-Calculated %RE Check

Goal: Fit a calibration curve on the analyte/IS response ratio and accept it by per-level back-calculation accuracy, not by R-squared.

Approach: Fit 1/x^2-weighted linear regression of response ratio on nominal concentration, back-calculate every standard, flag any non-LLOQ level outside +/-15% and the LLOQ outside +/-20%, and set the LLOQ to the lowest passing level.

standards <- data.frame(
  conc = c(1, 5, 10, 25, 50, 100, 250, 500, 1000),
  analyte_area = c(480, 2500, 4900, 12100, 24500, 49000, 121000, 245000, 488000),
  istd_area = c(100000, 98000, 99000, 101000, 100000, 98000, 101000, 99000, 100000)
)
standards$ratio <- standards$analyte_area / standards$istd_area

fit <- lm(ratio ~ conc, data = standards, weights = 1 / standards$conc^2)
standards$back_calc <- (standards$ratio - coef(fit)[1]) / coef(fit)[2]
standards$re_pct <- (standards$back_calc - standards$conc) / standards$conc * 100

# ICH M10: each calibrator within +-15%, +-20% at the LLOQ (lowest level)
tol <- ifelse(standards$conc == min(standards$conc), 20, 15)
standards$pass <- abs(standards$re_pct) <= tol
lloq <- min(standards$conc[standards$pass])

Internal-Standard Normalization

Goal: Convert raw analyte area to a matrix-corrected response that the calibration curve maps to concentration.

Approach: Divide analyte area by co-eluting SIL-IS area per sample, then invert the same response-ratio calibration; matrix effect and extraction recovery cancel in the ratio.

samples$ratio <- samples$analyte_area / samples$istd_area
samples$conc <- (samples$ratio - coef(fit)[1]) / coef(fit)[2]
samples$conc[samples$conc < lloq] <- NA   # below validated range -> not reportable

Ion-Ratio Confirmation

Goal: Guard against quantifying an isobaric co-eluter as the analyte.

Approach: Compute the qualifier/quantifier area ratio per sample, compare to the mean calibrator ratio, and flag samples outside the tolerance window -- a drifted ratio means the quantifier peak is partly something else.

cal_ratio <- mean(standards$qualifier_area / standards$quantifier_area)
samples$ion_ratio <- samples$qualifier_area / samples$quantifier_area
# SANTE/2020/12830 uses +-30% relative for LC-MS/MS qualifier/quantifier ratios
samples$id_confirmed <- abs(samples$ion_ratio - cal_ratio) / cal_ratio <= 0.30

MRM gives mass selectivity, not identity: two compounds can share a precursor->product transition (many acylcarnitines share m/z 85; lipids share head-group fragments). Identity needs retention time plus the ion ratio plus an authentic standard. Near the LLOQ the qualifier may fall below its own detection limit, so ion-ratio confirmation is usually only enforceable above a few times the LLOQ -- state that limit rather than hiding it. A single-transition method has no defense against isobaric interference and is a documented compromise, not a default.

LOD and LLOQ

Goal: Set the lowest reliably quantifiable concentration from noise and accuracy, not from an extrapolated curve.

Approach: Estimate LOD from blank-signal scatter (S/N ~3) and confirm the LLOQ as the lowest calibrator meeting the +/-20% back-calculation and precision criteria; never anchor the curve below the real noise floor to claim sensitivity.

blank_areas <- c(100, 120, 95, 110, 105)
slope <- coef(fit)[2]
lod <- (mean(blank_areas) + 3 * sd(blank_areas)) / slope   # S/N~3 convention
# LLOQ is the lowest calibrator passing +-20% %RE AND precision -- not 10*SD/slope alone

Per-Method Failure Modes

Matrix suppression unaccounted

Trigger: Neat-solvent calibration, or an IS that does not co-elute with the analyte.
Mechanism: Co-eluting phospholipids/salts suppress analyte ionization; with no co-eluting IS the suppression does not cancel.
Symptom: Solvent and matrix-matched curves disagree; IS-normalized matrix factor far from 1; lot-to-lot drift.
Fix: Co-eluting SIL-IS per analyte; map suppression zones by post-column infusion and move peaks off them; report IS-normalized matrix factor across at least six matrix lots.

One IS shared across chemically diverse analytes

Trigger: A single global IS used to correct a heterogeneous panel.
Mechanism: The IS corrects suppression and recovery only for analytes co-eluting and chemically near it; distant analytes carry the difference of two suppressions.
Symptom: Excellent CVs but biased group means -- precision (set by the IS correcting injection/drift) and accuracy (set by the per-analyte residual) are decoupled.
Fix: SIL-IS per analyte, or per RT/chemical cluster; treat low CV as no evidence of correctness.

Unweighted calibration over a wide range

Trigger: OLS fit on heteroscedastic data; acceptance judged by R-squared.
Mechanism: High-concentration points dominate least squares; the low end is fit poorly.
Symptom: R-squared 0.999 yet +30-40% bias at the LLOQ.
Fix: Compare 1/x and 1/x^2, pick by minimizing low-end |%RE|, judge by per-level back-calculation.

Isotopic crosstalk between analyte and IS

Trigger: IS-analyte mass gap below ~3-4 Da, or high IS:analyte ratio at the LLOQ.
Mechanism: The analyte natural-isotope envelope bleeds into the IS channel (bends the high end, mis-read as saturation); IS isotopic impurity bleeds into the analyte quantifier channel (inflates the low end, biases the LLOQ badly).
Symptom: Non-linear high end; a matrix-blank-plus-IS sample shows signal in the analyte channel.
Fix: Choose an IS mass gap of at least 3-4 Da or a less-abundant SIL isotopologue transition; always run a zero (matrix + IS only) and require its analyte-channel signal below 20% of the LLOQ.

Pre-analytical degradation

Trigger: Delayed quench, freeze-thaw, slow time-to-freezer.
Mechanism: Metabolism continues post-collection (glycolysis, esterases, redox auto-oxidation); labile metabolites collapse in seconds to minutes.
Symptom: No chromatographic error -- the true value is biased before injection; low/variable adenylate energy charge across samples is the tell that quenching, not biology, drove the numbers.
Fix: Cold (-40 to -80 C) aqueous-organic quench matched to the metabolite, measure freeze-thaw and long-term stability per labile analyte, control collection-to-freeze time as a study variable.

Quantitative Thresholds

Threshold	Source	Rationale
Calibrator back-calc within +/-15% (+/-20% at LLOQ), >=75% of >=6 levels pass	ICH M10 (Step 4, 2022)	Per-level accuracy, not correlation, defines a usable curve
QC accuracy +/-15% (+/-20% at LLOQ); precision CV <=15% (<=20% at LLOQ)	ICH M10	Intra- and inter-day acceptance at >=4 levels
IS-normalized matrix factor CV <=15% across >=6 lots	ICH M10 / Matuszewski 2003	Proof the IS cancels matrix effect; raw MF may be poor while IS-normalized MF ~1
Carryover <=20% of LLOQ (analyte), <=5% (IS)	ICH M10	Measured in a blank after the ULOQ; concentration-dependent, must be quantified not eyeballed
Selectivity: interference at LLOQ <=20% of analyte, <=5% of IS response	ICH M10	Across >=6 individual matrix lots
ISR: >=2/3 of reanalyzed study samples within +/-20%	ICH M10	Only test that catches incurred-sample-specific problems spiked QCs cannot
Ion-ratio tolerance +/-30% relative (LC-MS/MS)	SANTE/2020/12830	Illustrative codified window; enforce only above a few times the LLOQ
>=10-15 points across a chromatographic peak	Convention	Reliable integration; sets the cycle-time ceiling
S/N ~3 = LOD, ~5-10 = LLOQ	Convention	Detection vs reliable quantification; LLOQ also bounded by accuracy/precision

Common Errors

Error / symptom	Cause	Solution
Curve accepted on R-squared, biased at LLOQ	Unweighted heteroscedastic fit	Weight (1/x, 1/x^2); accept by per-level back-calculated %RE
Deuterated IS gives lot-dependent ratios	Deuterium isotope effect separates IS from analyte; lost matrix correction	Use 13C/15N at non-exchangeable positions, or verify co-elution explicitly
High-end curvature mis-read as detector saturation	Analyte natural isotopes bleed into a too-close IS channel	Widen IS-analyte mass gap to >=3-4 Da; use nonlinear isotopic-crosstalk correction
Beautiful CVs, wrong group means	One global IS across diverse analytes -- precision/accuracy decoupled	SIL-IS per analyte or per RT/chemical cluster
Low samples after a high sample read high	Concentration-dependent carryover	Inject a blank after the ULOQ, randomize run order, report measured carryover
Skyline never ratios analyte to IS	IS not tagged Label Type = heavy and paired to its light analyte	Set Label Type heavy in the transition list; pair by molecule name
Validated assay, study numbers still wrong	Pre-analytical degradation (no error message)	Quench fast, measure stability, monitor adenylate energy charge

References

MacLean B, Tomazela DM, Shulman N, Chambers M, Finney GL, Frewen B, Kern R, Tabb DL, Liebler DC, MacCoss MJ. 2010. Skyline: an open source document editor for creating and analyzing targeted proteomics experiments. Bioinformatics 26(7):966-968.
Peterson AC, Russell JD, Bailey DJ, Westphall MS, Coon JJ. 2012. Parallel reaction monitoring for high resolution and high mass accuracy quantitative, targeted proteomics. Molecular & Cellular Proteomics 11(11):1475-1488.
Matuszewski BK, Constanzer ML, Chavez-Eng CM. 2003. Strategies for the assessment of matrix effect in quantitative bioanalytical methods based on HPLC-MS/MS. Analytical Chemistry 75(13):3019-3030.
ICH M10 Bioanalytical Method Validation and Study Sample Analysis. ICH Harmonised Guideline, Step 4, adopted 24 May 2022 (FDA implemented November 2022; EMA effective January 2023).
Broadhurst D, Goodacre R, Reinke SN, Kuligowski J, Wilson ID, Lewis MR, Dunn WB. 2018. Guidelines and considerations for the use of system suitability and quality control samples in mass spectrometry assays applied in untargeted clinical metabolomic studies. Metabolomics 14(6):72.
Wang S, Cyronak M, Yang E. 2007. Does a stable isotopically labeled internal standard always correct analyte response? A matrix effect study on a LC/MS/MS method for the determination of carvedilol enantiomers in human plasma. Journal of Pharmaceutical and Biomedical Analysis 43(2):701-707.
Teo G, Chew WS, Burla BJ, Herr DR, Tai ES, Wenk MR, Torta F, Choi H. 2020. MRMkit: automated data processing for large-scale targeted metabolomics analysis. Analytical Chemistry 92(20):13677-13682.

Related Skills

metabolomics/xcms-preprocessing - Upstream feature detection for untargeted discovery before targeted validation
metabolomics/statistical-analysis - Group comparison and multivariate analysis of quantified concentrations
metabolomics/isotope-tracing - Stable-isotope tracing and flux (MID), the adjacent discipline this skill hands off to
metabolomics/normalization-qc - QC-sample-driven drift correction and RSD filtering
clinical-biostatistics/cdisc-data-handling - Regulated-trial bioanalysis data handling when targeted numbers feed a clinical study