Convert Log2 (FPKM+1) to FPKM in an expression matrix outputted from RSEM

Question

[dina.hesham139]

Hi, I have a processed file of online RNAseq data that contains log2(x+1) transformed gene-level expression values (FPKM) matrix. I don't know how to proceed with the analysis with that type of file. For instance, I tried to proceed with DEseq2 but I got this Error:

Error in DESeqDataSet(se, design = design, ignoreRank) : some values in assay are negative

So, I think that I should revert the Log2 (FPKM+1) to FPKM to remove the negative values. Any idea how?

If there's another way to proceed with the file as is please enlighten me :)

this is the header of the file:

gene_Symbol CTNNB1_WT1 CTNNB1_WT2 CTNNB1_WT3 CTNNB1_WT4 CTNNB1_WT5 CTNNB1_WT6 CTNNB1_Mutant1 CTNNB1_Mutant2 1 CLU 11.62513 13.26637 9.137255 7.133718 11.87641 10.956011 11.01832 12.42982 2 EEF1A1 11.29064 10.95692 11.354335 11.210096 11.90874 11.776091 12.23826 11.46498 3 FTH1 11.43760 11.99286 10.118945 11.614278 10.25135 11.595979 10.77482 11.52371 4 ACTB 11.11159 11.67700 9.995405 10.422961 10.06434 10.799942 10.25373 10.73077 5 GFAP 12.55514 13.99373 4.239152 2.931744 12.02517 3.777809 11.45016 13.79807 6 SPARC 10.00989 10.88951 8.801095 8.682285 11.83168 9.901304 10.62619 11.35472

Thanks in advance!

Answer 1

[kelen]

If you are planning to use DESeq2 for DE, I would advise against using FPKM values - read more about that in the DESeq2 manual. I would also suggest you refresh your knowledge on how to inverse logarithms - it is useful to know these (e.g. https://mathinsight.org/logarithm_basics just the first page that popped up when googling). Lastly, in case I am missing something obvious then having a negative value after log2(FPKM+1) raises a red flag for me.

[dina.hesham139]

Thank you for your reply! I agree I can't imagine a reason why there are negative values after log2(FPKM+1).

[kelen]

You say in the title that this is an expression matrix from RSEM - if you can get your hands on the actual output from RSEM, then those results can easily be used in DESeq2 (via tximport()).

[dina.hesham139]

Unfortunately, I don't have the actual RSEM files. I will try to request them.

Answer 2

[Hamid Ghaedi]

Both Deseq2 and edgeR assumed that you are providing raw count and not any form of processed data like FPKM and log2(FPKM). So you can not use these methods to determine differentially expressed genes by using FPKM or log2(FPKM).

[dina.hesham139]

Are there any other pipelines you would recommend to proceed with?

[ATpoint]

limma-trend, please use the search function, this has been asked dozens of times before.