Skip to content

conditionTest() yields mostly 0 and 1 p-values (non-normal distribution) #45

@QINYIXIANu

Description

@QINYIXIANu

I’m following the protocol from the KRAS vignette ([https://hectorrdb.github.io/condimentsPaper/articles/KRAS.html])
My dataset consists of one shared lineage (Lineage1) across two conditions, with approximately 200k nuclei per condition.
I fitted the model using nknots = 6, and ran:
condRes <- conditionTest(kras, l2fc = log2(2), lineages = TRUE)
condRes$padj <- p.adjust(condRes$pvalue, "fdr")
All parameters are left as default. However, I noticed that my resulting p-values are mostly either 0 or 1. There’s no smooth or approximately normal distribution.

Image

Is this expected behavior, or could it indicate a model fitting or design issue?
I wonder whether this could be due to overfitting with nknots = 6, unbalanced pseudotime coverage between 2 conditions, or some other parameter choice.

Any insights on how to interpret this p-value distribution, or diagnostics I could check would be very helpful.
Thank you!

Best,
Yixian Qin

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions