Contrast p-value are different from default-provided p-values

[MaximilianNuber]

Dear all,

Thank you for this great package.
I was testing how to apply the contrasts you describe in your vignette to my data.
I have a categorical covariate with 3 levels, one of them the reference (let´s call that level 1).
To my understanding, the logFCs and values provided in the results for level 2 and 3 respectively, should be in comparison to level 1 as the intercept.
Now I applied the contrast from level 2 to level 1, the intercept, and the p-values were either 0 or close to 0, when the default p-values were very mixed.
I would trust the default values primarily, but is it expected for the result to be different?

Thank you for any help.

Best,
Max

[lhe17]

Hi MaximilianNuber, Thank you for your question. "To my understanding, the logFCs and values provided in the results for level 2 and 3 respectively, should be in comparison to level 1 as the intercept." Yes, this is correct. "Now I applied the contrast from level 2 to level 1, the intercept" The problem here is that you compare the logFC, which is the difference between level 2 and level 1, with the intercept, which is the mean expression under level 1. You should not expect these two quantities to be the same. So you should use the default value (logFC of level 2). In addition, the intercept reported in the output of nebula is the intercept term after all covariates are standardized (i.e., centered and scaled). I will add this information to the manual and vignettes. Best regards, Liang

…

On Sun, Mar 17, 2024 at 4:59 PM MaximilianNuber ***@***.***> wrote: Dear all, Thank you for this great package. I was testing how to apply the contrasts you describe in your vignette to my data. I have a categorical covariate with 3 levels, one of them the reference (let´s call that level 1). To my understanding, the logFCs and values provided in the results for level 2 and 3 respectively, should be in comparison to level 1 as the intercept. Now I applied the contrast from level 2 to level 1, the intercept, and the p-values were either 0 or close to 0, when the default p-values were very mixed. I would trust the default values primarily, but is it expected for the result to be different? Thank you for any help. Best, Max — Reply to this email directly, view it on GitHub <#46>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGDISURRAXZGX6ZABBSFJXLYYX7ZTAVCNFSM6AAAAABE2QBWOKVHI2DSMVQWIX3LMV43ASLTON2WKOZSGE4TAOBWGI3DCNY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>