This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Can I convert normalized counts from DESEQ2 into fpkm?

Hi all, So I got gene expression values from different RNA-seq experiments. Some are FPKM and some are normalized counts from DESEQ2 I believe I cannot compare these directly. Is there any way to make them comparable? Maybe a way to convert normalized deseq counts to FPKM values?

Thanks

rna-seq deseq normalization fpkm
• 6,435 views • link
written by nanoide nanoide

1 answer

Do you have access to the raw counts? The point with DESeq2 counts is that the size factors change depending on which samples you process together. Should start with raw counts of all experiments if possible.

• 1 views • link
written by ATpoint ATpoint

That's what I thought. Unfortunately, these are not my studies and I doubt I can get the raw counts. If i did, I should use DESeqDataSetFromHTSeqCount() and then fpkm(), rigth?

Thank you for your tiime

• 1 views • link
written by nanoide nanoide

Yes, but I would avoid FPKM as literature states on multiple occasions that it performs poorly on between-sample comparisons. Check literature on comparisons of normalization methods. I would go with the standard DESeq2 workflow, getting normalized counts from that pipeline.

• 1 views • link
written by ATpoint ATpoint

By default, DESeq2's fpkm() will return "robust" FPKMs, which use size factors to normalize instead of taking the column sums of the raw counts.

• 1 views • link
written by igor igor

But this would mean that FPKMs are only comparable if all the samples were processed together and contributed to the size factors, right?

• 1 views • link
written by ATpoint ATpoint

Thanks for the useful discussion. Then this means I should use robust = F in fpkm() in deseq? So I'm able to directly compare with other fpkm values? Thanks

• 1 views • link
written by nanoide nanoide

Non-robust should be the classic FPKMs. If you check the column sums (colSums), they should all add to 1M then.

• 1 views • link
written by igor igor

Yes. But if the samples are processed separately, you will likely have batch differences. That's a much more complex issue.

• 1 views • link
written by igor igor

Log in to answer this question.

More posts like this