I am confused about RNASeq quantification and normalization. There seems to be debate on the correct way to normalize RNASeq data, and the creator of RPKM/FPKM now advocates using TPM instead (http://liorpachter.wordpress.com/201...t/#comment-246). However, Cuffdiff seems to be the only differential expression tool that uses normalized expression values (I may be wrong here), but it is not clear to me exactly how Cuffdiff works. In contrast, count based differential expression methods (DESeq, edgeR, EBSeq and others) try to use counts and do not focus of TPM vs. FPKM issues of normalization.
My question is, what is the point of RNASeq normalization if the majority of differential expression methods do not even use these normalization methods? Is the normalization for comparison of the general "expression level" of a single gene across multiple samples, or possibly to understand whether a gene is highly or lowly expressed relative to other genes within a sample? Alternatively, should differential expression tools attempt to use normalized values?
My question is, what is the point of RNASeq normalization if the majority of differential expression methods do not even use these normalization methods? Is the normalization for comparison of the general "expression level" of a single gene across multiple samples, or possibly to understand whether a gene is highly or lowly expressed relative to other genes within a sample? Alternatively, should differential expression tools attempt to use normalized values?
Comment