Error Bars Following Normalisation Of Real Time Pcr Data

Question

[Dolores Hamilton]

Using the 2-delta delta Ct method I have determined the relative gene expression in a panel of 7 treated and untreated cell lines-part of data shown below (C=untreated; T=treated):

Sample RQ        RQ        RQ
D556C  2.6588333 2.1409998 4.9944763
D556T  9.368949  7.3571334 18.470724
D384C  14.2748   10.199899 18.23931
D384T  11.659178 9.1894222 7.546831

I used Prism to represent the data as a bar chart with error bars

For each cell line I want to look at the fold change in expression between the untreated and treated, and so for each replicate and cell line I normalised the treated to the untreated-part of data shown below:

Sample    Foldchange     Foldchange   Foldchange
D556C       1                1           1
D556T       3.5              3.4         3.7

Using prism when I graphically represent the fold change data I get error bars for the treated cell line, how do I get error bars for the untreated cell line to which the treated has been normalised? i.e all untreated replicates have a value of 1. When I look in the literature results are displayed with error bars on the untreated 1 x sample?

Many thanks

Dolores

[Neilfws]

This is not very clear at all. Are you saying that there are 7 cell lines, each either treated or untreated? It would help if you could show some raw data and explain what software was used. It also sounds as though you have 2 observations per cell line (1 treated, 1 untreated), in which case error bars are quite meaningless.

Answer 1

[Chris Evelo]

Supposing you have normalized the treated samples to the average of the untreated samples for the same cell line you can simply calculate the standard deviation between the untreated samples and normalize that to the average as well, and then represent the result as the relative error. If your samples are somehow paired (i.e. one specific untreated sample was used as the control for one specific treated sample), you can only do the normalization on a 1 by 1 basis (which is actually what you did in the example data that you now added), and it makes no sense to show the variation among the control samples, since these in that case cannot be normalized themselves.