Skip to contents

Wrapper function to generate module membership parameters

params_module_membership.Rd

Controls how a gene x k loading matrix from ICA, NMF or DGRDL is turned into module membership. Genes are kept where they sit in the tail of a component's loading distribution, which means membership is not exclusive: a gene loading strongly on three components belongs to three modules. Genes in no tail belong to nothing, which is the background category an argmax assignment cannot give you.

Usage

params_module_membership(
  method = c("zscore", "fdr"),
  cutoff = 3,
  fdr = 0.05,
  tails = c("auto", "upper", "both"),
  scaling = c("robust", "standard")
)

Arguments

method

String. "zscore" standardises each component and keeps abs(z) > cutoff. "fdr" converts to two-sided p-values against a Normal null fitted the same way, Benjamini-Hochberg adjusts, and keeps padj < fdr. One of c("zscore", "fdr"). Defaults to "zscore".

cutoff

Numeric. Absolute z threshold for method = "zscore". Defaults to 3.0.

fdr

Numeric. Adjusted p-value threshold for method = "fdr". Defaults to 0.05.

tails

String. "auto" uses an upper-tail-only test when every loading is non-negative (the NMF case) and a two-sided one otherwise. "upper" and "both" force the choice. One of c("auto", "upper", "both"). Defaults to "auto".

scaling

String. "robust" centres and scales each component by its median and MAD. "standard" uses the mean and standard deviation instead, which is stricter and less forgiving of skewed loadings (e.g. NMF). One of c("robust", "standard"). Defaults to "robust".

Value

A named list with the following elements:

  • method - String. "zscore" standardises each component and keeps abs(z) > cutoff. "fdr" converts to two-sided p-values against a Normal null fitted the same way, Benjamini-Hochberg adjusts, and keeps padj < fdr. One of c("zscore", "fdr"). Defaults to "zscore".

  • cutoff - Numeric. Absolute z threshold for method = "zscore". Defaults to 3.0.

  • fdr - Numeric. Adjusted p-value threshold for method = "fdr". Defaults to 0.05.

  • tails - String. "auto" uses an upper-tail-only test when every loading is non-negative (the NMF case) and a two-sided one otherwise. "upper" and "both" force the choice. One of c("auto", "upper", "both"). Defaults to "auto".

  • scaling - String. "robust" centres and scales each component by its median and MAD. "standard" uses the mean and standard deviation instead, which is stricter and less forgiving of skewed loadings (e.g. NMF). One of c("robust", "standard"). Defaults to "robust".

References

Biton, et al., Cell Rep, 2014