The attached graph shows the expression of a protein (Y-axis) at each cell division (X-axis) in three different treatment groups.
I would like to compare the red and orange groups (test) to the blue (control) not at each individual cell division but taking into account the whole data set.
Line represents mean and shade represents SD. n=3
In GraphPad Prism I chose the two-way ANOVA test and in the multiple comparison tab chose the Dunnett's test. This test is recommended when comparing multiple groups to a control group and, in this particular case, takes into account the man of the whole data set for each group.
Is this the appropriate statistical test for the question that I am trying to answer: is red or orange different from blue?
"Is this the appropriate statistical test for the question that I am trying to answer: is red or orange different from blue?"
No, not exactly, because you don't compare the curves - you only compare the expression at each time point. Further, the data between different time-points of the same group are not independent.
Ideally, you should model the protein concentration as a function of time (cell divisions). It looks like a simple saturation function. You can the compare biologically relevant parameters of the kinetic, e.g. if the initial production rate is different, or if the asmytotic value is different between the groups.
I noticed that the expression at t=0 is negative. How can this be / is this correct? Does the plot show log expression?
https://www.graphpad.com/support/faqid/1554/
Hi Jochen and thanks for the reply.
So the X-axis is not time point but cell division # but I guess your point is the same nonetheless. Indeed the data for each replicate is associated across the cell division span.
Thanks for the modelling idea. I will try that!
As for the negative values, this is data derived from flow cytometry and compensation can sometimes cause a parameter to register negative values after the mathematical corrections.
The unit of the time variable is not relevant for modelling (if it is hours, days, cell divisions, ...). It's only important to interpret the rate coefficient of the model that will have the corresponding unit (per day or per cell division ...).
Having negative values then is a sign that the correction method is not well chosen. That's typically obtained from a simple subtraction. Some convolution model might be better, but in your case I think that would be shooting sparrows with cannons (so just keep in the back of your mind that the correction might be suboptimal).
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA