Median-centred log-ratio with optional MAD scaling

Robust descriptive transform on log-abundances: subtracts the per-sample median log-abundance (more robust to outlier features than mean-centering as in CLR/rCLR) and optionally divides by the per-sample median absolute deviation (MAD; limits leverage from individual high-magnitude features).

This is NOT an Aitchison-valid simplex projection. It is a robust descriptive transform on log-abundances, suitable as input to learners that do not require the zero-sum compositional coordinate constraint.

Usage

ws_mad_logratio(x, pseudocount = 0.5, scale_by_mad = TRUE)

Arguments

x

Samples x features matrix (rows = samples).

pseudocount

Additive pseudocount before logging. Default 0.5.

scale_by_mad

Logical. If TRUE, divides each sample by its per-sample MAD after median-centering. Default TRUE.

Value

Same shape as x; the median-centred (MAD-scaled) log-abundance values, with attributes centering and scaling.