Fisher's Exact Test Across All Cell Types and pSI Thresholds

fisher.iteration tests a candidate gene list for over-representation among the genes specific to each cell type/sample, at each of the four standard pSI thresholds (0.05, 0.01, 0.001 and 0.0001), using a one-sided Fisher's exact test.

Usage

fisher.iteration(
  pSIs,
  candidate.genes,
  background = "data.set",
  p.adjust = TRUE
)

Arguments

pSIs: data frame output from specificity.index, with one column per cell type/sample and genes as rows.
candidate.genes: character vector of official gene symbols to test for over-representation.
background: character string naming the background gene list for the test. The default "data.set" uses all genes in pSIs; use it when the candidate genes and the pSI data set come from the same species. "human.mouse" restricts the background to the genes in pSIs (human) that have a clear mouse homolog, for testing a mouse candidate list against a human pSI data set; "mouse.human" is the converse. These two options download the UCSC kgXref table for mm10 or hg19 (a file of several megabytes) from https://hgdownload.soe.ucsc.edu/ and stop with an informative error if the download fails.
p.adjust: logical. If TRUE (the default) p-values are adjusted for multiple testing across cell types with the Benjamini-Hochberg method (p.adjust, method "BH"), separately for each threshold; if FALSE, nominal p-values are returned.

Value

A data frame with one row per cell type/sample and four columns, one per pSI threshold (0.05, 0.01, 0.001, 0.0001). Column names end in "adjusted" or "nominal" depending on p.adjust.

Details

For every cell type/threshold combination the function asks: what is the probability that this many candidate genes would be specific to this cell type by chance? Each gene in the background is classified by two binary variables, specific to the cell type at this threshold or not, and in the candidate list or not, and the resulting 2 x 2 table is tested with fisher.test (alternative "greater"). Gene symbols are compared in upper case, and duplicated candidate genes are counted once.

Author

Xiaoxiao Xu, Alan B. Wells, David R. O'Brien, Arye Nehorai, Joseph D. Dougherty

Examples

##load sample pSI output
data(sample.data)
##load sample candidate gene lists
data(candidate.genes)
##run Fisher's exact test for overrepresentation on pSI.out for the AutDB
##candidate gene list across all cell types/sample types & pSI thresholds
fisher.out.AutDB <- fisher.iteration(pSIs=sample.data$pSI.output,
                                     candidate.genes=candidate.genes$AutDB)
fisher.out.AutDB
#>                             0.05 - adjusted 0.01 - adjusted 0.001 - adjusted
#> Amygdala.Young.Adulthood          1.0000000       1.0000000        1.0000000
#> Cerebellum.Young.Adulthood        0.1782927       0.5452343        0.1298184
#> Cortex.Young.Adulthood            0.2663041       0.5452343        1.0000000
#> Hippocampus.Young.Adulthood       0.6336715       0.5452343        0.2348364
#> Striatum.Young.Adulthood          0.7711528       0.5452343        0.2348364
#> Thalamus.Young.Adulthood          0.4268605       0.5452343        0.2348364
#>                             1e-04 - adjusted
#> Amygdala.Young.Adulthood                   1
#> Cerebellum.Young.Adulthood                 1
#> Cortex.Young.Adulthood                     1
#> Hippocampus.Young.Adulthood                1
#> Striatum.Young.Adulthood                   1
#> Thalamus.Young.Adulthood                   1