Inverse-log preprocessor for pre-log-transformed microarray deposits

Converts a matrix of log-transformed expression intensities back to a positive abundance-like scale via \(base^x\). This is required when a public deposit has already applied a \(\log_2\) (or \(\log_{10}\)) transform before archival (common in Toray 3D-Gene and some Agilent deposits). The resulting matrix is on a positive scale comparable to raw NGS counts and can be fed directly to within-sample Module A methods (rCLR, ILR, ALR-pivot, MAD log-ratio, dominance score) which require positive compositional input.

Non-finite values (Inf, -Inf, NaN) are replaced with the per-row median of finite values, with a warning.

Usage

preprocess_inverse_log(x, base = 2, pseudo = 0)

Arguments

x: Numeric matrix (features \(\times\) samples, or samples \(\times\) features — the function operates element-wise). Both orientations are accepted; the transform is purely element-wise.
base: Numeric. The log base used by the deposit. Default 2 (Toray 3D-Gene standard). Use 10 for log10 deposits.
pseudo: Numeric. Small offset added to the output to keep strictly positive values. Default 0 (output of \(base^x\) is always \(> 0\) for finite inputs).

Value

Numeric matrix of the same dimensions as x, on a positive abundance scale. The attribute preprocessing records the transform applied.

References

Toray Industries, Inc. (3D-Gene microRNA chip) — data format specification.

Examples

if (FALSE) { # \dontrun{
set.seed(42)
log2_matrix <- matrix(rnorm(50 * 20), nrow = 50)
colnames(log2_matrix) <- paste0("S", seq_len(20))
rownames(log2_matrix) <- paste0("miR-", seq_len(50))
abundance <- preprocess_inverse_log(log2_matrix, base = 2)
stopifnot(all(abundance > 0))
} # }