B09: Reusable teaching module
Distinguish analytical features from confirmed metabolites; evaluate internal standards and scaling; separate concentration changes from relative signal differences.
Metabolomics measures small molecules using platforms with different coverage and selectivity. A detected feature may describe a mass/charge and retention-time signal rather than a confidently identified compound. Adducts, isotopes and fragments can yield multiple features associated with one molecule. Conversely, co-eluting compounds can complicate a signal.
Targeted assays can quantify defined compounds with calibrations; untargeted assays often yield relative intensities with varying identification confidence. Processing should document blanks, technical QC samples, drift, batches, detection limits and normalization. A table heading called concentration does not turn an arbitrary instrument signal into an absolute chemical measurement.
If the observed signal is \(x_{ij}=a_j c_{ij}\) and an internal standard has known concentration \(c_s\) and the same multiplicative factor \(a_j\), the ratio \(x_{ij}/s_j\) cancels that factor. This works only to the extent that the standard tracks the relevant measurement process.
Scaling a feature to mean zero and variance one changes its contribution to a multivariate analysis. It does not convert arbitrary units into comparable concentrations.
The inputs below are hypothetical. Run the example, predict one change, then compare your prediction with the result. All executable code uses base R.
concentration signal standard ratio
1 2 1.6 4.0 0.4
2 3 3.3 5.5 0.6
3 4 5.6 7.0 0.8The common drift cancels in this deliberately idealised example. If the analyte and standard respond differently to preparation or the instrument, that cancellation will be incomplete.
A chromatographic mass-spectrometry feature is often defined by mass-to-charge ratio, retention time and signal. Isotopes, adducts and fragments can create several features from one compound. An accurate mass alone can match multiple candidate structures. Supporting evidence can include fragmentation and comparison with standards under documented conditions.
Maintain an identification field separately from an abundance field. A feature can be quantified reproducibly while its chemical identity remains uncertain. Conversely, a confidently identified compound can have imprecise abundance measurements. Do not turn candidate identities into definite pathway membership when interpreting a differential-feature list.
Targeted concentration requires a calibration model and an applicable range. Relative untargeted intensity can support within-feature comparisons but is not automatically comparable in molar units across compounds. Extraction recovery and ionization differ by molecule and sample matrix.
Consider a simple hypothetical response \(S=a+bC\), with intercept \(a\) and sensitivity \(b\). Estimating concentration from signal assumes that calibration applies to the analysed samples and that the relation remains appropriate over the range used.
calibration <- data.frame(concentration=c(0,1,2,4,8),
signal=c(3,13,23,43,83))
cal <- lm(signal~concentration,calibration)
unknown_signal <- c(18,63,123)
estimated <- (unknown_signal-coef(cal)[1])/coef(cal)[2]
data.frame(signal=unknown_signal,estimated_concentration=estimated,
within_calibration=estimated>=0 & estimated<=8) signal estimated_concentration within_calibration
1 18 1.5 TRUE
2 63 6.0 TRUE
3 123 12.0 FALSEThe last estimate extrapolates beyond the standards. Perfect linearity in this constructed example does not justify extrapolation in a real assay. Calibration uncertainty, blanks, detection limits and replicate measurements still need evaluation.
An internal standard can track selected preparation or instrument variation. A pooled QC sample can reveal drift across acquisition order. A blank can reveal background or carryover. None alone represents independent biological replication, and none automatically diagnoses all compounds equally.
If acquisition order follows treatment, drift can imitate a group effect. Distributing biological groups across the run and recording acquisition order supplies diagnostic information. A correction should be based on a justified technical model; excessive correction can remove biology if the technical and biological factors are confounded.
Log transformations emphasize multiplicative changes. Unit-variance scaling makes features contribute equal marginal variance to a multivariate analysis. A rare, noisy feature can therefore gain influence. Other scaling rules balance this trade-off differently; the choice is part of the analysis rather than a cosmetic plotting option.
high_signal low_signal
original_variance 71.27785 0.000111737
scaled_variance 1.00000 1.000000000The equal scaled variances say nothing about equal biological information or chemical concentration. Inspect reliability and missingness before deciding which features enter PCA or prediction.
Three correlated features receive the same candidate compound name. Are they three independent confirmations of pathway change? They may be adducts or related signals from one compound. Examine their analytical relationship and identification evidence before counting biological support.
Chemical identity should have documented supporting evidence; a feature ID is not a metabolic pathway assignment. Avoid counting several signals from the same compound as independent confirmations. See the EMBL-EBI metabolite-identification lesson for the distinction between a detected signal and an assigned compound.
EMBL-EBI introduction to metabolomics provides broader context. Read it alongside the distinction between relative feature intensity and calibrated concentration in this module.
Metabolomics processing, exploratory analysis and integration.
Use the unified map to locate B09 in both harmonized outlines. Return to the building-block index.