How to provide the fold change value of a group of biological replicates? I want to visualize relative gene expression with barplot + error bars after qPCR analysis. I was asked to use the log2FoldChange values for the barplots, where each bar should summarize 3 values (the biological replicates).
I calculate the delta delta CT (ddCT) as dCT(treated) - dCT(control).
The fold change is calculated as 2^ddCT.
From which value can I calculate the mean for the representative value of all three replicates (and should I take arithmetic or geometric mean)? Should I take the average of the ddCTs first and then exponentiate it for Fold change? Or can I take the average of the 3 fold changes?




condition
ddCT
Fold change




untreated
0
1


treatment1
-0.4997
0.7072


treatment2
-3.8894
0.0674


untreated
0
1


treatment1
1.00437
2.0060


treatment2
-1.485
0.3570


untreated
0
1


treatment1
-2.0324
0.2444


treatment2
-0.4631
0.7253



 A: Quantitative polymerase chain reaction (qPCR) data are initially reported as the number of cycles, $C_T$, needed for a specific PCR-amplified nucleic-acid product to exceed a threshold quantity. Delta-$C_T$ (d$C_T$) values represent differences in $C_T$ between the analyte of interest and an analyte thought to be independent of the experimental conditions (like a "housekeeping" gene), to try to control for differences in starting material between samples. Delta-delta-$C_T$ (dd$C_T$) values can be differences in d$C_T$ between experimental treatments and a control, used to report the effects of the treatments on the analyte of interest.
You generally want to do calculations of means and estimate confidence intervals in a scale where errors are fairly symmetrically distributed above and below the mean. From that perspective it's best to work in original units related to $C_T$ values. The exponentiation needed to translate $C_T$-related values to fold changes can lead to very large skew. If you want to report confidence levels in terms of fold change, first calculate the confidence levels in the $C_T$-related scales and only exponentiate at the end.
Although you have used your control condition as a basis with dd$C_T$ of 0, don't forget to show error bars based on its d$C_T$ values, to give your audience an estimate of reproducibility. Also evaluate whether the control nucleic acid used to get the d$C_T$ values is itself unaffected by your experimental treatments. Otherwise, apparent results on your analyte of interest might instead represent effects of your experimental manipulations on that control analyte.
