We currently have a study done with RNA-seq analysis. But the raw counts in our data show a large difference, up to 10 fold higher in the raw counts versus the lowest one. But after normalization, it shows 1.2 fold changes.
Is this 10-fold change in raw counts generally acceptable in the field of RNA-seq?
Hi Shenq-Shyang Huang , I'm assuming you're referring to a difference in the raw counts of a single gene/location across replicates? If so, the normalization should (in theory) adjust for these differences using "size factor" scaling. That said, you may want to do some careful quality control to make sure there are no quality issues with the sequencing and/or alignment.
Hi Mark Pepin , thank you for the answer. We currently have over 100 samples taken from human tissues, and we've checked the quality (such as RIN, 260/280 before RNA-seq). There are some samples that show high raw counts 10-fold higher compared to the lowest one. And I am curious given that the normalization step gives us low changes, does that mean the quality is acceptable? or do we need to check other parameters while doing RNA-seq? We used STAR to align and analyzed the quality by MultiQC.
Dear Shenq.
An additional quality control step after normalization is to check the densities plots with log10 normalized expression values to compare the distributions from all libraries. Overlapping expression distributions indicate appropriate homogeneity of the sequencing depth and that count normalization was suitable to compare the expression levels of the different libraries. Non-overlapping densities indicate that comparison of libraries would not be appropriated, you can check if maybe some libraries have excess of PCR duplication of reads. For more details on this analysis an as an example you can check my publication
Article Transcriptome profiling at osmotic and ionic phases of salt ...
Hope this helps
Hi Diana,
Thank you for the reference and suggestion, that is a good idea for us to check the quality of our samples.
Hello Shenq-Shyang Huang, Generally in RNA_Seq analysis after alignment it is important to remove duplicates in bam files (picard is a good tool for this). After the quantification process, your counts must follow a pre-processing analysis and visualize after normalization (PCA, heatmap or density curve). These will tell you if these differences are due to technical bias or not. Je ne sais pas quel outil vous avez utiliser pour les analyses statistiques (edge,deseq2...) I don't know what tool you used for the statistical analyzes (edge, deseq2...) you will also have to refer to the manual of the tools you used for these pre-processing analyzes.
- Badges
- Science method