I have performed differential expressed genes (DEGs) using the DESeq2 package. I understood that this package has its own normalization method to calculate for DEGs. To represent the data on gene expression value, is it proper to use the normalized DESeq2 value or is it better to use FPKM/TPM value instead?
Yes you can. The values you obtain would be the normalized counts, and this should reflect the expression level of the gene.
If you are making the comparison between different conditions for the same gene, then a higher normalized count would indicate higher expression.
However if you need to make comparisons across genes, for example you would want to claim that some genes are more expressed than others, for the same condition, then converting to FPKM/TPM would work better
You should use the DEG from DESeq2 if you are comparing the same gene across different samples. DESeq2 normalization help to get more statistical power by doing its own normalization but that normalization do not take gene length into account and you should not perform analysis that requiers that type of data or your should keep in mind and advise that your data contain bias for long genes. (Note that in the latest version of DESeq2, the log2FolChange shrinkage function have been separated from the DESeq function and that you should pick one of the 3 option available depending of you data and run it after the DESeq function. Shrinkage allows a better ranking and visualization of the data without the use of an arbitrary filtering of the gene with low counts.)
If you want to perform enrichment analysis, I highly suggest you using goseq to weight the DEG list and the term list with the gene length. You can use goseq with any library that you want if you format it the right way.
And if you want to perform tests that require homoskedatic data (PCA, clustering), you should use the rlog or the vst transformation of DESeq2 .
- Badges
- Science method