There are rna-seq data for the test sample and control sample that treated the drug. I would like to proceed from bam file to count data, deg analysis, and pathway analysis.
Can I check the quality of these samples and make an evaluation that some samples are not suitable for analysis (even though they are test samples, they are classified similar to control) after analysis?
Is it the difference between the data qc and the sample qc considering the experimental conditions... Can someone you know explain it to me.
Have you considering doing a standard limma analysis and simply allowing limma to downweight any outlier samples? For example, voomLmFit() with sample.weights=TRUE.
Any test sample that is very different from the other test samples with automatically get downweighted, meaning that you don't need to make arbitrary decisions yourself.
If you have test samples clustering with controls, then that would seem to be a biological effect, and there is no way that a technical QC analysis can diagnose a biological effect. Technical data problems would generally cause an outlier sample rather than tranforming a test sample into a control.
I did the DEG analysis with the default option of edgeR or DESeq. I'll have to learn more about limma analysis. Then, it seems to be likely that the sample was not a problem in the sequencing stage, but in the sampling or experimental stage. Thank you for the detailed explanation!
Log in to answer this question.
I don't understand, please explain better. What I get is that you have samples, but what do you mean by QC? Like, enough reads, and things like that?
I wrote a reply below. :)
So your test samples look like controls. Based on what kind of analysis? PCA?
Data QC will only look at the quality of sequence data (e.g. with a program like FastQC). You will be evaluating quality score and basecall distribution etc. This part does not consider experimental conditions.
Sample QC is going to look at the unsupervised clustering of samples (based on the entire data for each) to see if the samples cluster near each other (according to sample type and/or experimental condition being tested). You could also be looking for presence/absence of expression of specific genes/group of genes based on prior knowledge and your expectations/expertise about the organism being studied.
Thank you. I understood about data qc, but I've never done sample qc in detail. It can be also verified by the expression of gene sets through background knowledge. Since the data has already been analyzed once before, I got prior information that one test sample is located close to the control cluster through hierarchical clustering, but I didn't understand it well.