I basically need some advice with regards to which statistical test I should use to determine whether there is a statistically significant difference between my data. Briefly my experiment involves using a line measure to measure the fluorescence intensity across cell-cell junctions from a fluorescence microscopy micrograph. I measure at a particular time point at 10 different junctions in each condition and it produces graphs such as the one attached (see link). The graphs produce fairly different shapes, due to the fact that the cell-cell junctions become broader and more variable in the HDM condition, however an unpaired t-test does not reflect this. I thinking that possibly the data is not evenly distributed and therefore I am better off doing a non-parametric test instead? Any help would be greatly appreciated. https://ibb.co/mFcTAG
There is much more in your data than you can address with t-tests or U-tests. The picure shows that the fluorescence peaks as a specific distance. There are several possibly relevant differences between the groups:
- the distance at which the peaks occur,
- the peak heights (max fluorescence),
- the width of the peaks
If these are relevant things, then I suggest to use a non-linear model to describe the peaks (and their differences depending on the group). I'd say that a simple model of the form a*exp(((x-b)/c)²) + d will be ok, where a=peak height, b=peak position, c=peak width, and d=baseline value (offset).
Hi Gabriel,
The non-parametric test corresponding to the unpaired t test is Mann- whitney U test.
You can go through it, that is best applicable in this.
But before aplying this check the normality of both the group (by any of kolmogorov-smrinov test or shapiro wilk test or shapiro francia test or box plot), if any of the group is non-normal then go for mann- whitney test, but if both the group are normal then you should stay with unpaired t test.
Thanks
There is much more in your data than you can address with t-tests or U-tests. The picure shows that the fluorescence peaks as a specific distance. There are several possibly relevant differences between the groups:
- the distance at which the peaks occur,
- the peak heights (max fluorescence),
- the width of the peaks
If these are relevant things, then I suggest to use a non-linear model to describe the peaks (and their differences depending on the group). I'd say that a simple model of the form a*exp(((x-b)/c)²) + d will be ok, where a=peak height, b=peak position, c=peak width, and d=baseline value (offset).
I think that the junction distance (JD) is producing a common impact on (X,Y)=(PBS, HDM). So if normality holds, you can use multivariate multiple regression of (X,Y) on JD.
See the slides no. 13 onwards in the following link:
http://www.public.iastate.edu/~maitra/stat501/lectures/MultivariateRegression.pdf.
Alternately, you can use Functional ANOVA with PBS and HDM as two groups where each one is a function of a continuous variable JD.
Hi thanks for all of your help. In relation to the non-linear model suggested, the curves reflect relative fluorescence ie they have been normalised to 1, therefore do you think peak heights is relevant? Obviously if the max fluorescence (1) is occurring at different x values (junction distance) for the junctions of a particular condition, this reduces the peak height, as opposed to the max fluoro occurring consistently at the same x value.
Also I need to take away background from the graphs. Is the best way to do this to take a line scan of the same length from the area proximal to each junction and take this away from the junction line scan values?
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs