Entering edit mode

3 months ago

alwayshope
▴
40

Dear experts,

When analyzing the bulk RNAseq data, using DESeq2, the log2FC value distribution is close to a normal distribution, while the t value distribution is not like a normal distribution.

While if use limma, the test value and the log2FC distribution are all similar to a normal distribution,

So, does the difference caused by the heterogeneity of the samples themselves or the algorithms of the two tools?

Thank you very much!

did you apply one of the variance stablization algorithms to the data? if so the two will be incongruous.

Thanks a lot!

The results of DESeq2 and Limma are similar, the majority are overlapped.

The reason for this question is that when trying to use the decoupleR package to infer the pathway activities,

it uses t or the statistics values to infer the pathway activity, and the distribution of the t vlaues is close to a normal distribution, Using log2FC seems can give similar results;

While for my results, DESeq2 log2FC follows the normal distribution, not it's statistics values, so I use log2FC for the input of decoupleR instead.

Thanks a lot!

Which test did you use in DESeq2, Wald-test or likelihood ratio test (LRT)?

(....) For the Wald test, stat is the Wald statistic: the log2FoldChange divided by lfcSE, which is compared to a standard Normal distribution to generate a two-tailed pvalue. For the likelihood ratio test (LRT), stat is the difference in deviance between the reduced model and the full model, which is compared to a chi-squared distribution to generate a pvalue.

Thanks a lot!

I used DEseq2 and the likelihood ratio test (LRT) to get the log2FC and the test values and use these for the decoupleR.

Best,

There comes out a new question for the usage of decoupleR, the distribution of my counts after log-transformed does not follow the normal distribution, so can I still use decoupleR for this batch of my data?

Are you taking log2 of raw counts? It looks like you just need normalized counts from DESeq2 and stats for each genes for

`decoupleR`

.Yes, it is the log2 of raw counts.

I will try to use the normalized counts from DESeq2 and stats for each gene for decoupleR.

Thank you very much for your kind guidance!

Best,

alwayshope - for the future, this follow up was different enough to where we would have encouraged you to

1) check if this question has already been asked?

2) post it as a separate question if not

3) link back to this question for context

Thanks a lot, will follow your guidance! And post a new question if too different of the follow up questions.

Thank you very much!

Best regards,