Question

How are the P and adjusted P values calculated in de_analysis() from Lipidr?

0

Entering edit mode

yoyoStat • 0

@yoyostat-22954

Last seen 4.7 years ago

The package help mentioned that it doesn't use ordinary t-test. I checked the code of de_analysis(), and it seems that it's passing data to lmFit(). The result of de_analysis() includes P-value and adjust P-value. How are they calculated? And can I use this to check if the concentration of a lipid species is changed significantly between 2 groups of samples?

BTW, my input data is absolute lipid concentration (nmol/10million cells). Do I have to normalize the data using normalize_pqn()? Could anyone tell me under what circumstances should normalization be used?

lipidr • 908 views

ADD COMMENT • link updated 4.7 years ago by ahmed.mohamed ▴ 10 • written 4.8 years ago by yoyoStat • 0

1

Entering edit mode

I have removed the "limma" tag. Although lipidr uses limma under the hood, I can't answer questions about lipidr or what it includes as output.

ADD REPLY • link 4.8 years ago Gordon Smyth 52k

1

Entering edit mode

If you check the source code of de-analysis.R line 135, you see it used topTable, so type ?topTable to learn about its parameters. Since de-analysis.R does not seem to give any other than the default parameters, adj.p should represent the default BH-corrected p-values. The lines before 135 tell you how the differential analysis was performed (standard limma workflow).

ADD REPLY • link 4.8 years ago ATpoint ★ 4.5k

score 1 · Answer 1 · 2020-03-11

Sorry for the late answer, I don't usually check the support site.

As mentioned in the comments, the output of lipidr::de_analysis is basically a limma::topTable, which reports the P values and adjusted P values using "BH" method by default.

PQN normalization adjusts for unknown dilution errors that may affect the overall lipid concentration. Assuming that you already calculated your absolute lipid concentration using internal standards, or a similar correction method, you don't need to redo the normalization using PQN.

Hope this helps.