Hello!
I have a question regarding the quantification of data obtained from microscopy. It is clear that if you take several images from the same organ/tissue, you can not consider one picture as an independent data. Therefore, one should take several pictures from the same organ/tissue, make an average, and the consider averages from different organs/tissues as a independent samples.
My question comes when you are using data from cells. Every cell is different for the other. When one is quantifying, for example, number of particles/organelles, morphological parameters or colocalization, how can we plot the data? Usually I try to analyze more than 50 cells in 3 independent experiments. The problem comes in the analysis, if I do the average of each replicate and then I plot only 3 values, sometimes the differences are not significative, maybe because they are not so big difference. If I plot all the values in 3 different columns, then small changes become significative.
Does anybody know which is the proper way? I think a t-test is not powerful enough to consider hierarchical data. Does anybody know any other test in which it can be considered all the data but taking into account that it is hierarchical data and not independent samples?
Thanks for your answers!
Hi, I'm honestly not so skilled in statistics, but maybe Tukye's range test might be the test you're looking for. It works well in conjunction with the anova and it simply tells you which of your experiments significantly differ from each other.
Hi David,
Averaging the measurements from each set of cells is one way of doing it, and t-test is a good way to determine whether the differences are statistically significant. From your description it is not clear what you measure, but 50 cells is a relatively low number of cells. With such a low number you should expect problems getting statistical significance, but you could increase the number of cells to get a stronger statistical significance in the t-test.
The other way of doing that is by using pattern recognition. You could use CellProfiler or Wndchrm to do your analysis. Wndchrm can also determine the hierarchy in the form of a phylogeny.
My back ground is remote sensing (satellite and other types of imaging of the earth surface including the bottom of the ocean) and something that I have used with the digital images I work with to see differences is:
1. Analyze not only the average but also the variance within the data sets (perhaps plot the std dev of the data sets?)
2. If you apply principal component analyses (PCA) using only two data sets as input at a time similar things will go into principal component number 1 and things that are different will go into component number 2. The variance of component number 2 will give you some information as to how alike or different the images are from each other.
3. Use spatial variability analyses to look at textural and spatial pattern differences between the images.
Since I come from a different field the terms and words used may not be familiar; I do have an interest in the application of the remote sensing expertise to the medical imaging field.
Pat
- Badges
- Science method