Linear Models for Differential Expression Analysis using limma

Perform differential expression analysis using linear models with empirical Bayes moderation from the limma package. Supports both two-group and multi-group comparisons.

Usage

gly_limma(
  exp,
  group_col = "group",
  covariate_cols = NULL,
  subject_col = NULL,
  p_adj_method = "BH",
  ref_group = NULL,
  contrasts = NULL,
  add_info = TRUE,
  ...,
  log_transform = TRUE
)

Arguments

exp: A glyexp::GlycomicSE() or glyexp::GlycoproteomicSE() object, or another SummarizedExperiment containing expression data and sample information.
group_col: A character string specifying the column name in sample information that contains group labels. Default is "group".
covariate_cols: A character vector specifying column names in sample information to include as covariates in the limma model. Default is NULL.
subject_col: A character string specifying the column name in sample information that contains subject identifiers for paired comparisons. Default is NULL.
p_adj_method: A character string specifying the method for multiple testing correction. Must be one of the methods supported by stats::p.adjust(). Default is "BH" (Benjamini-Hochberg). Set to NULL to skip p-value adjustment.
ref_group: A character string specifying the reference group. If NULL (default), the first level of the group factor is used as the reference. Only used for two-group comparisons.
contrasts: A character vector specifying custom contrasts. If NULL (default), all pairwise comparisons are automatically generated, and the levels coming first in the factor will be used as the reference group. Supports two formats: "group1-group2" or "group1_vs_group2". Use the second format if group names contain hyphens. "group1" will be used as the reference group.
add_info: A logical value. If TRUE (default), variable information from the experiment will be added to the result tibble. If FALSE, only the statistical results are returned.
...: Additional arguments passed to limma::lmFit().
log_transform: A single non-missing logical value. If TRUE (default), transform expression values using log2(x + 1e-6) before analysis. If FALSE, use the supplied values unchanged, for example for already log-transformed data.

Value

A list with three elements:

tidy_result: A tibble with limma results containing the following columns:
- variable: Variable name
- log2fc: Model contrast on the analysis scale (log2 fold change when expression values are on a log2 scale)
- AveExpr: Average expression level
- t: t-statistic
- p_val: Raw p-value
- p_adj: Adjusted p-value (if p_adj_method is not NULL)
- b: B-statistic (log-odds of differential expression)
- ref_group: Reference group
- test_group: Test group
raw_result: The raw limma fit object(s).
meta_data: A list containing metadata from the input experiment.

Details

By default, expression data are transformed using log2(x + 1e-6) before statistical testing. Set log_transform = FALSE to skip this transformation. The analysis uses linear models with empirical Bayes moderation to improve statistical power, especially for small sample sizes.

The log2fc column always contains the fitted model contrast. When log_transform = FALSE, its units are those of the supplied expression data; it is a log2 fold change only when those values are already on a log2 scale.

For two or more groups, all pairwise comparisons are automatically generated (one contrast for two groups) and performed using contrast matrices unless custom contrasts are specified.

If covariates are provided, they are added as additional terms in the design matrix and are not part of the group contrasts.

If subjects are provided, they are included as a blocking factor in the design matrix using the formula ~ 0 + group + subject (plus any covariates).

When specifying custom contrasts, use either "group1-group2" or "group1_vs_group2" format. If group names contain hyphens and you use the first format, an error will be raised suggesting to use the second format.

Required packages

This function requires the following packages to be installed:

limma for linear model fitting and empirical Bayes moderation