Meta-analyse cell-type composition shifts across studies

Scores each sample for each cell type as the mean within-study z-score of log-CPM expression across that cell type's marker genes (an MCP-counter style abundance proxy), computes Hedges' g for case vs control in every study, and combines the per-study effects with a DerSimonian-Laird random-effects model. A significant shift indicates that part of the bulk differential expression may reflect changes in tissue composition.

Usage

mx_cell_composition(studies, markers, min_markers = 3)

Arguments

studies: A named list of metaXpressStudy objects.
markers: A named list mapping cell types to marker gene identifiers, or a data.frame with columns gene and cell_type. Identifiers must match rownames of the studies.
min_markers: Minimum number of expressed markers a cell type needs in a study to be scored there. Default 3.

Value

A data.frame sorted by p-value with columns cell_type, meta_g (pooled Hedges' g; positive = higher in cases), pvalue, padj, i_squared, n_studies, and one g_<study> column per study. Per-sample scores are stored in attr(x, "scores").

Details

Expression is log2-CPM. If a study's colData has a lib_size column it is used as the library size, so studies that carry only a subset of genes are still normalised to their true depth.

References

Becht, E. et al. (2016) Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression. Genome Biology, 17, 218. doi:10.1186/s13059-016-1070-5

Hedges, L. V. (1981) Distribution theory for Glass's estimator of effect size and related estimators. Journal of Educational Statistics, 6(2), 107–128. doi:10.3102/10769986006002107

Examples

if (FALSE) { # \dontrun{
  markers <- list(T_cells = c("CD3D", "CD3E", "CD2"),
                  Fibroblasts = c("COL1A1", "COL3A1", "DCN"))
  comp <- mx_cell_composition(studies, markers)
} # }