Entering edit mode
6.8 years ago
firatuyulur
▴
320
Hi, I am given a set of RNASeq fastqs to carry out diff. expression analysis. I have 3 replicates for both control and treatments. I use bcbio for the pipeline that takes care of many steps. Aligner is tophat2. I used DESeq2 for the diff exp part. The results are not very satisfying that in wet lab part the treatment's phenotype is very obvious on proliferation rates yet in deseq2 output the log2fc ranges between 0.5 to -0.4. If I take padj 0.05, there are only 2 significantly enriched genes in total. This does not feel right with such significant phenotype. Has anyone ever come across something like this?
There could be many reasons why you don't detect any link between gene expression changes and phenotype. It's all in the details which you don't give. Some reasons can be biological, others technical. Among possibilities are:
- the treatments affect the phenotype independently of change in gene expression
- the phenotype is related to changes in gene expression that you don't find significant with your test, e.g. the phenotype leads to noisy data (i.e. high variability) and also consider that small changes can sometimes have big biological effects (don't confound statistical significance and biological relevance)
- if the phenotype is not measured in the same samples used to extract the RNAs, it could be that the treatment didn't work in (some of) the RNA extraction sample(s)
- bugs in your code like instructing DESeq2 to do the wrong comparison
On the other hand, maybe you should be glad to only find two genes. Usually people find way more genes than they can make sense of.