Working with paired-end RNA-seq data for a diploid plant, both root and shoot separately, I currently have a matrix of read counts for a couple of thousand genes for both. I am now trying to identify genes differentially expressed in (e.g.) root with respect to shoot.
I have spent some time experimenting with RPKM and TPM values, but these don't really tell me much. Also tried EdgeR, and I have a smear-plot (attached), but am at a loss as to what this is really telling me.
My questions:
1. Why is EdgeR preferred over RPKM/TPM values?
2. What decides the threshold to be set for identifying truely differentially expressed genes?
3. If I use HTSeq rather than a simple script to count mapped reads, does it take into account only a small portion of the gene, or is its complete length mapped?
The problem with differential expression tests can be related to data normalization. Have you tried to normalize for the number of cells in each run?
Cheers.
Hi Omar,
EdgeR use raw count instead of RPKM. Length bias is also a problem in RNAseq data and RPKM use the length for the calculation of counts. Look at this pulication
http://www.ncbi.nlm.nih.gov/pubmed/18516045
I will suggest you to go for htseq-count for counting map reads and once you have the table you can use either DeSeq or EdgeR. Htseq-count returns only the number if reads mapped to gene.
For differentially expressed genes use FDR or BH multiple hypothesis correction at 0.05 or 0.001[ Whatever you want to use].
Dear Reema,
I already have a read count out of the filtered SAM file, which counts only reads mapped to each gene using a Perl script. I don't see how that would be any different from was htseq would give me. I have used exactly this count table to generate the smear plots using edgeR, and not RPKM values.
Hi Roop,
Becuase of length bias in RNAseq data EdgeR perfer raw count over RPKM values. If your perl script generating raw count then it's fine. Along with reads mapped to a gene/exon, Htseq-count also take into consideration other factors- ambigous reads, multiple aligned reads etc. But I hope you must have consider that in your perl script.
Best Wishes
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Bioinformatics
- Bioinformatic Tools