Bio Metabolomics Targeted Analysis

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

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

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

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.