Combines the results of several hit-finding methods in one table. For each gene it reports the FDR of each method and the largest of them (MaxFDR), a conservative score to rank the genes supported by all the methods. To get a complete picture, compute the tables with fdr = 1 so that they contain all the tested genes.
Arguments
- hit_list
A named list of tables, each with the columns Gene and FDR (for example from
find_gene_zscore_hit,find_camera_hit,find_roast_hitorfind_time_hit). The names are used in the column names of the result.- fdr
The FDR threshold used for the columns NMethods and Hit (default 0.05).
Value
A tibble with one row for each gene and the columns Gene, FDR_<method> for each method, NMethods (number of methods in which the gene has FDR <= fdr), MaxFDR and Hit (MaxFDR <= fdr), ordered by MaxFDR.
Details
Since an FDR (Benjamini-Hochberg adjusted p-value) is never smaller than the p-value, MaxFDR <= alpha implies that the gene has a p-value <= alpha in every method: for each single gene this is a level-alpha intersection-union test of the hypothesis "the gene has no effect in at least one method", valid whatever the dependence between the methods. Note however that the list of genes with MaxFDR <= alpha does not have a guaranteed False Discovery Rate: the false discoveries of the single methods can concentrate in their intersection.
A gene missing from a table gets an NA FDR for that method and an NA MaxFDR.
Examples
zscore <- data.frame(Gene = c("A", "B", "C"), FDR = c(0.01, 0.2, 0.03))
camera <- data.frame(Gene = c("A", "B", "C"), FDR = c(0.02, 0.01, 0.5))
summarise_hits(list(zscore = zscore, camera = camera))
#> # A tibble: 3 × 6
#> Gene FDR_zscore FDR_camera NMethods MaxFDR Hit
#> <chr> <dbl> <dbl> <dbl> <dbl> <lgl>
#> 1 A 0.01 0.02 2 0.02 TRUE
#> 2 B 0.2 0.01 1 0.2 FALSE
#> 3 C 0.03 0.5 1 0.5 FALSE