Skip to contents

The Sim2 Classification Dataset

Source: R/data.R
sim2.Rd

A simulated two-group classification dataset for demonstrating fit_mdm() and benchmark(). It contains 100 simulated 1D NMR spectra split evenly into groups A and B, where 3 out of 25 peaks per spectrum differ between the groups: in group A, two peaks are scaled by 1.25 and one peak by 1/1.25 (\(\approx 0.80\)). Group B is left unmodified. The first spectrum (sim2_001) is constructed without any global or per-peak ppm jitter so it can serve as a clean unshifted alignment reference.

Usage

sim2

Format

A spectra object consisting of 100 spectrum objects, where each spectrum contains 2048 datapoints ranging from 3.59 to 3.28 ppm. The per-spectrum group labels are attached as attr(sim2, "group"), a named factor with levels A and B. For details about spectrum and spectra objects see metabodeconplus-classes.

Each spectrum's meta$simpar carries the standard fields (x0, A, lambda, noise) plus five sim2-specific fields: base_x0 (the 25 reference peak positions, identical across spectra), dx0 (per-peak jitter in ppm), gx0 (scalar global ppm shift), diff_AB (integer indices into base_x0 of the peaks that differ between groups), and ab_factors (the multiplicative factors applied to those peaks in group A). They satisfy x0[k] = base_x0[k] + dx0[k] + gx0.

attr(sim2, "true_x0") is a numeric vector with the post-alignment ppm positions of the discriminating peaks (one per diff_AB index), useful for highlighting them on plots of aligned feature matrices.

Details

Peak parameters (positions, areas, half-widths and noise) were chosen to match the values recovered by deconvoluting the sim dataset, which itself is derived from the Blood reference dataset (see sim). Concretely:

  • 25 base peaks per spectrum with positions drawn uniformly in [3.37, 3.52] ppm. The reference-grid step is 0.00015 ppm/dp.

  • Per-peak jitter with standard deviation 4 datapoints (\(\approx 0.00060\) ppm) plus a per-spectrum global ppm shift with standard deviation 8 datapoints (\(\approx 0.00120\) ppm) to mimic chemical shift variation between samples.

  • Base areas drawn from a log-normal distribution centered around 2500 (in ppm-area units) and varied per spectrum by +/-60%.

  • Base half-widths drawn uniformly in [0.0009, 0.0013] ppm and varied per spectrum by +/-10%.

  • Gaussian noise with standard deviation 2200.

  • In group A, three base peaks (indices simpar$diff_AB) have their areas multiplied by c(1.25, 1.25, 1/1.25) (stored in simpar$ab_factors). Group B is left unmodified. Spectrum sim2_001 is generated with dx0 = 0 and gx0 = 0 to provide a clean unshifted alignment reference.