How does one determine whether a fold change calculated on qPCR data using 2-ΔΔCt method is significant? Answers provided to a similar question earlier indicate that one should do a one sided t-test on logFC. How does one go about this? Are there other methods to determine whether the FC is significant? Does one need to report the significance when reporting fold change? Your suggestions will be highly appreciated.
The logFC can be tested with "standrad methods" like the t-test.
The decision between one- and two-sided depends on what direction of regulation you would find interesting. If you want to find out if a gene is regulated and either direction would be a "hit", then a two-sided test is in place. If you use the test only to confirm a regulation in a particular direction, then a "statistically significant" regulation in the "wrong" direction is considered a "false positive", so in this case the null hypothesis will never be rejected (and thus never be wrongly rejected) when the direction is wrong. Therfore, the nominal type-I error rate will be kept with a one-sided test.
Are there other methods? To test the expected value of the logFC? No. Not really. Only the t-test does this (and a randomization or bootstrap test on the mean logFC). Other tests test different hypotheses. One can think of plenty different kinds of hypotheses, and each one will have at least one test (or one can design a test for the purpose).
Does one need to report the significance? Well, why did you do the test? I suppose that has some rationale... and if so, one should expect that the appropriate result is reported. According to Neyman, reporting the decision is enough. According to Fisher, the author should provide at least the p-value (actually also the statistic and the degrees of freedom).
If the test is not central, it is usually good to provide the confidence interval for the estimated logFC (or for the FC, if you wish).
A problem arises when there is left-censoring so you have values below the detection limit at baseline. A method has been recently developed for estimating the fold change in this situation, see the paper by Xu et al (2015) Estimation of intervention effect using
paired interval-censored data with clumping below lower detection limit. Statist. Med. 2015, 34 307–316
Paul Milligan
- Badges
- Science method