qRT-PCR analysis : average triplicates for statistical testing

Question

[Nicolas Rosewick]

Hi,

I've qRT-PCR data to analyze. Two cell lines, each cell lines treated with 5µL and 10µL treatment. Each treatment was perfomed in each cell lines twice, and triplicates were performed in the qRT-PCR experiments. Thus for each cell line and treatment I have two triplicates. For example for cell line A here are the different sample analyzed in qRT-PCR. ( e.g. 5µL_treatment.1 and 5µL_treatment.2 are the same treatment but done in different tubes)

Sample   Treatment
A        Mock
A        Mock
A        Mock
A        5µL_treatment.1
A        5µL_treatment.1
A        5µL_treatment.1
A        5µL_treatment.2
A        5µL_treatment.2
A        5µL_treatment.2
A        10µL_treatment.1
A        10µL_treatment.1
A        10µL_treatment.1
A        10µL_treatment.2
A        10µL_treatment.2
A        10µL_treatment.2

I've performed 2^ddct analysis and now want to perform a stat test (Mann-Whitney). Do I have to average the triplicate before doing the test or should I use all technical replicates in my analysis.

Thanks

Answer 1

[Devon Ryan]

You need to do the averaging before the test. Otherwise you'd be lying to yourself about how much statistical power you had.

BTW, unless you're hitting a ceiling/floor effect you might use a T-test instead (though only having two replicates limits what you'll be able to detect).

[Nicolas Rosewick]

Ok thanks that's what I thought

[Nicolas Rosewick]

you recommend a t-test for qRT-PCR analysis ? on 2^ddct, or dCTs ?

[Devon Ryan]

I usually just used the dCt values. The Mann-Whitney test is mostly useful when you start hitting the Ct detection limit. Then you really need a non-parametric, otherwise you'll have better power with a T-test (unless of course the normality assumption is sufficiently violated...but with so few samples you'll have to just assume that).