model p values , adjusted or not ?

[kiwipeel]

Hi,

What are the p values we get after running run_ulm, consensus , mlm etc. ? Are those p adjusted or not ?

# question

[PauBadiaM]

Hi @kiwipeel,
The p-values obtained depend on the method. For ulm and mlm, it is the p-value of the coefficient fitted in the linear model, for the consensus score it is the mean p-value across methods. In all cases p-values are not corrected, we leave that to the user.

[kiwipeel]

Hi @PauBadiaM . Thank you for your answer.

I am doing psedobulk pathway activity analysis and use the DEGs I got after doing condition A_celltypeX vs. Condition B_celltypeX. So I am running the codes for each cell type seperalty and get p values for each celltype.

Sould I do p value correction on p values obtained from each cell type results or should I combine all results, all p values in a dataframe and then to the p value adjustment?

Which p value correction method is best for correction consensus p values?

I am using the code below

results <- decouple(
  mat = deg,           ###loop code uses different DEGs or each celltype 
  network = net,
  .source = "source",
  .target = "target",
  statistics = c( "wsum", "ulm","mlm"),
  args = list(
    wsum = list(.mor = "weight", times = 1000,.likelihood = "likelihood"),
    ulm = list(.mor = "weight",.likelihood = "likelihood"),
    mlm = list(.mor = "weight",.likelihood = "likelihood")
  ),consensus_stats = c("mlm", "ulm" , "norm_wsum"),
  consensus_score = TRUE,
  minsize = 5
)

contrast_acts  <- subset(results, subset = results$statistic == "consensus" )

[PauBadiaM]

Hi @kiwipeel,

Sorry for the delayed response. Personally I would just use ulm instead of the consensus, as it can underperform if there is co-linearity in your net as mlm scores will be noisier. I would correct p-values inside each cell-type contrast, using the BH correction method but any other would be acceptable. Hope this is helpful!