Filter multi-omics feature blocks before RF fitting

Usage

filter_omics(
  dat.list,
  filter_mode = c("auto", "none", "manual"),
  filter_method = c("mad", "variance"),
  top_n_by_type = NULL,
  top_n_manual = NULL,
  return_summary = TRUE,
  verbose = TRUE,
  ...
)

Arguments

dat.list: A named list containing multi-omics datasets with samples in rows and features in columns.
filter_mode: Feature filtering mode. "auto" applies omics-aware defaults, "none" keeps all features, and "manual" uses top_n_manual.
filter_method: Feature dispersion metric used for ranking. Choose from "mad" or "variance".
top_n_by_type: Optional named numeric/list object used in "auto" mode. Supported names are rna, mirna, methylation, protein, and unknown. Defaults are RNA = 5000, miRNA = adaptive 500-1000, methylation = adaptive 5000-20000, protein = all, unknown = all.
top_n_manual: Optional named or positional numeric/list object used in "manual" mode. Names should match dat.list block names.
return_summary: Logical; whether to return filtering summary metadata.
verbose: Logical; whether to print per-block filtering messages.
...: Additional arguments (currently ignored).

Value

If return_summary = FALSE, returns filtered dat.list. Otherwise returns a list with dat_filtered and filter_summary.

Details

Before feature ranking, all blocks are checked to have the same samples in the same order, unique sample names, numeric finite values, and at least two samples. Zero-variance features are removed before optional scaling or RF fitting because they cannot define a split and would become non-finite under z-standardization. Column names are preserved verbatim.