What are the data values in qRT-PCR for statistical testing? dCT? or 2^-dCT?

Question

[YCW121]

Hi, I read a lot of papers and I found that they didn't indicate what kind of values are used in t-test or logistic regression. Here is part of my data. The miR-29b-3p is the target gene and RNU48 is the internal control in my study. The dCT is calculated by CT of miR-29b-3p minus CT of RNU48.

I want to find differentially expressed genes between patients with good treatment responses and patients with poor treatment responses. So, should I use dCT or 2^-dCT to perform the t-test and logistic regression to predict their response? I built a logistic regression model and I found that the AUC is smaller if I use 2^-dCT to compute. But, if I use dCT, the AUC increased from 0.65 to 0.75. I don't know which is correct. Correct me if I am wrong. Please help me, thank you.

Answer 1

[karl.stamm]

Most statistical tests come down to, what are you really looking for?

dCT is an artifact of the instrument and subject to experimental biases, 2^-dCT is a conversion of that in a logarithmic way. Can you get to the units of measure you really care about? You said you want to find differentially expressed genes, but what does that mean?

More or less mRNA in the sample, in terms of nanograms, I suppose, is proportional to 2^-dCT, so that's the better path. But if you can get an udnerstanding of what you're measuring, that's ideal.

Answer 2

[dsull]

Ct is on a log2 scale. Ct of 7 vs. Ct 4 means that there were 3 cycles in-between and therefore 8x more nucleic acids (log2 of 8 = 3). Do you care about a log2 fold change of 3 or an 8-fold change?

My argument is: Do not use 2^-dCt. Why? Your data will not be distributed correctly. A dCt of zero means no change (2^-0 = 1). A dCt of 20 means 2^-20 = 9.53e-7. A dCt of -20 means 2^-(-20) = 1048576.

Do you see the problem now? Whenever dCt > 0 (change in one direction), your 2^-x is scrunched between 0 and 1. Whenever dCt < 0 (change in the opposite direction), your 2^-x falls between 1 and infinity. That's not symmetric and is certainly nowhere near a normal distribution.

If you insist on using a 2^-x transformation, please use a nonparametric statistical test.