I am working with RNAseq and qRT-PCR validation results to see a correlation. I am comparing a diseased group to a healthy group of cells. I have 4 biological replicates in each category. However, the qRT-PCR data I have is with 3 technical replicates for each of those categories. I haven't done any biological replicates so far. I read from somewhere that the dCt values are comparable to -log (FPKM) or -log (RPKM) and not ddCt. My first question is whether the qRT-PCR results without the biological replicates can be comparable to the RNAseq data having different biological replicates? The next question is, if I do biological replicates in qRT-PCR validation, then it will ddCt values comparing between different groups and how can that be correlated with FPKM or RPKMs? Also, I would like to know how the FPKM/RPKM can be calculated other than using DESeq2 or R. Are there any other ways? Plus if we have to use the normal way where after calculating the RPM, divide that by gene length, which gene length should be considered since there can be varying lengths of transcripts (and the RNAseq was done for the total RNA)?
Thanks in advance!
I would recommend against trying to compare FPKM against values found using qRT-PCR. Generally, qRT-PCR should be done to validate the gene expression changes you observe using RNAseq (i.e., does the experimental ddCT value agree with the experimental FPKM / control FPKM value). Plotting the FPKM values of each group should look similar to the 2-dCT values of those same samples, but I don't think that the correlation between these values would be the same for all genes.
Ideally, you should use the same samples that you used for RNAseq to do the qRT-PCR validation. I'd also be very clear on what you mean by "technical replicates for each of those categories." Does this mean you've done qRT-PCR on one sample from each category? If so, you will need to do qRT-PCR on additional samples to make statistical comparisons (remember, a qRT-PCR experiment gives you 1 value per sample/gene combination no matter how many wells you use to measure it).
A few other tools can do read count normalization, but DESeq2, edgeR, and FeatureCounts are among the most commonly used.
Normalizing read counts to gene length is done using the length of each gene individually.
-dCt is linearily related to to log(FPKM).
ddCt is linearily related to log(FPKM[treated]/FPKM[control]).
I do not recomment to exponentiate (d)dCt values. Better use the log of FPKM values. These variables ((d)dCt and log(FPKM)) have better statistical properties like symmetric distributions and homogeneous variances).
In contrast to John's suggestion I do not think that it is or that it should be required to use the same samples for validation. If you use the same samples, then it's just a technical validation (you confirmed that both techniques will give similar results on the same sample - but this is trivia and well known). More important is to show that a new sample behaves as expected. And if only a couple of genes are interesting, it can be more efficient to use qPCR instead of an independent sequencing experiment.
If the targeted sequences are from coding RNA, an even better validation would be to show (or investigate) the regulation on protein level (IHC, Western blot, mass spec). And finally to show the relevance of the regulated genes for the biological response (overexpression, knock-down).
If you show, for instance, that the protein X is induced and (1) a conditional overexpression of X leads to the same observed biological response and (2) blocking X in treated organisms/cells is preventing or reverting the response to the treatment is the ultimate validation. After seeing this, who cares if you could show the gene regulation with a couple of different techniques? And on the other hand, showing the regulation of the gene with different techniques (usually) remains scientifically irrelevant as long as you cannot demonstrate that this has some (meaningful) biological effect.
Of course can you compare results obtained with different sample sizes. If you have a couple of genes, you can analyze the correlation between the experiments (Seq and qPCR) across these genes. If you have n>1, you take the mean, otherwise you take the one value you have (which is, logically, also the mean of a single value). If you want to check a single gene, you need some information about the precision of your estimates. This is typically obtained from (biological) replications. You can then do a confirmatory hypothesis test using the data from the validation experiment.
Thank you very much for all your comments and thoughts on this. It helps a lot. A quick question. Is it possible to calculate the FPKM manually? It's about dividing by million and then by the gene length right? Because I have only particular genes that need FPKM. In that case, how do we identify the gene length from varying lengths of transcripts?
FPKM is not about a gene, it's about a transcript. There can be different transcripts produced from one gene.
https://bioinformatics.stackexchange.com/questions/66/how-to-compute-rpkm-in-r
- Badges
- Science method