I have 26 individuals from whose we extracted the microbiota over 20 times or so per individual. So, after cleaning and filtering the data, we ended up with 50 different OUTs and we conducted a LSA (Local Similarity Analysis) to see the correlation between the OTUs in each individual separately.
We saw that the correlation matrices were more or less equal between each other, and I conducted also a pairwise Mantel test to see if there were statistical differences between the subjects. The results shown that they were the same kind of correlation matrix.
So, my question is this: we want to join all the information into one matrix to make easier the interpretation and visualization of the data. I did the mean of the values, but I think that with this method I'm loosing valuable information. Do you know how it can be done properly?
Regards.
maybe its better to freshly calculate correlation based on groups instead of calculating metrics of metrics. Maybe this gives inspiration: http://stats.stackexchange.com/questions/4040/r-compute-correlation-by-group
If the matrices are essentially similar (as your Mantel tests suggest), then you probably aren't losing much information by using the mean values for each cell. Just be sure to explain the method in your paper and emphasize that the matrix you're showing is more along lines of an illustration of "typical" findings, rather than a specific result for a sample. In a way, what you're doing is a kind of meta-analysis, isn't it? Maybe you can call it that.
Thanks a lot to both of you. I think that I'll do what you have said, Stephen.
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology
- Medical Pharmacology
- Clinical Pharmacology