Hi, I am trying to do gene set enrichment using MAST. I mostly follow the vignette. Here is the code:
gseaFDR <- 0.05 bootstrapReplicates <- 3# 3 for temp testing - takes too long module <- "BTM" # seems, the only one they have anyway packageExt <- system.file("extdata", package='MAST') module_file <- list.files(packageExt, pattern = module, full.names = TRUE) gene_set <- getGmt(module_file) # getting GeneSetCollection gene_ids <- geneIds(gene_set) # extracting gene IDs print(head(gene_ids)) ## gene sets for human genes; to compare with mouse, remove all capitalization sets_indices <- limma::ids2indices(sapply(gene_ids,tolower), tolower(mcols(scaObj)$symbol)) print(head(sets_indices)) start_time <- Sys.time() boots <- bootVcov1(zlmCond, R=bootstrapReplicates) end_time <- Sys.time() message("\n bootstrapping time ") print(end_time - start_time) gsea <- gseaAfterBoot(zlmCond, boots, sets_indices, CoefficientHypothesis("conditionKO")) z_stat_comb <- summary(gsea, testType='normal') sigModules <- z_stat_comb[combined_adj<gseaFDR]
When I run the code with 3 repetitions of bootstrapping (
bootstrapReplicates <- 3), everything looks fine, but when I run it for real, with 60 bootstrapReplicates, I get a number (7 to be exact) of such warnings:
In .bayesglm.fit.loop.printWarnings(Warning, state, family) : fitted probabilities numerically 0 or 1 occurred
So, I have a few questions:
- Can I trust the fit and consequently the enrichment?
- What are these warnings talking about?
- What, if anything, to do about them?