Hello,

It is my first time doing RNA-Seq data analysis and I don't have any experience in this field. I would highly appreciate if I could get answers for the following questions that I did not understand during my data analysis. I have used 4 bio-replicates for treatment and control each. After following the pipeline to process data, I got FPKM value from cufflink. My questions are:

1. When I looked at the raw FPKM value across the sample bio-replicates for ~1700 genes, I noticed that very few genes have FPKM value for all 4 bio-replicates (not having 0 value) and rest of the genes have FPKM values with 0 for 2/3 replicates. I am not sure how I can account for that FPKM value variance among sample replicates and which steps I should follow for the quality assessment of the data?

2. I used the FPKM value to calculate the differential expression usinf DESeq package in R. But I got P value for all most all the up-regulated (>1.5 fold) and down-regulated (>-1.5 fold) genes in a range of 0.5-0.9. The P value is not <0.05. The adjusted P value is 1 for all of them. Does it mean that those deferentially expressed genes are not statistically significant? How and what is the way to calculate statistical significance of deferentially expressed genes in DESeq?

3. Is that variance among bio-replicates caused this P value in my analysis?  

4. Is another differential expression analysis such as Limma or EdgeR better to account for the variability among bio- replicates?

1. Uh, I wouldn't use cufflinks nor FPKM. Lior Pachter doesn't do that anymore and cufflinks comes from his lab, so you might consider using something more modern like salmon or kallisto and importing using tximport. As for the zeros, with only 1700 genes that might be expected? I suppose it depends on the genes and the tissue.
2. You shouldn't use FPKM with any package that is intended for count data. This includes DESeq, DESeq2, edgeR, and probably several others that escape me now.
3. Maybe. Or maybe you don't have enough depth for cufflinks to be useful in this context. Pseudo-alignments to the transcriptome are probably the way to go here - see #1.
4. Differential expression of RNA-seq data using limma and voom().
    

Hi,

Thank you for your reply and suggestions. I am wondering that after aligning with reference genome using TopHat, can I use Salmon or Kallisto that you suggested for transcript assembly? Should I use the raw numbers from cufflink that I currently have (before converting to FPKM value) to calculate differential expression using Limma?

I am confused with so many options for differential expression such as DESeq, Limma or edge R. Which one is the most ideal, reliable and commonly used package for RNA Seq data analysis? 

Thanks again.