I'm currently conducting a 16S microbiome analysis on 18 mice that were sampled at five time points. They were divided into three groups of six and given either a MOG+CFA emulsion, only CFA, or were left untreated. There was a single time point for each mouse taken before treatment administration.
The idea was floated by my PI that I should lump all observations across treatments into a single pre-treatment group and that when I do any statistical testing, differential abundance analysis, etc. where the treatment group is taken into account that for time=0, ALL mice are included in this group and considered either "CFA" "CTRL" or "MOG".
My question is if this is even a statistically sound decision? I've been given the argument that since the mice are all pre-treatment that they can be considered equivalent. But since we are doing a time series analysis, I feel like adding all mice into each treatment group would add an additional source of variation to anything I do. Subsequent time points depend on the corresponding time points from the previous individual and we could be introducing cohort specific effects with this in anything we do. Are there any takes on this approach?
Hi David,
I would recommend keeping your groups separate despite these data being pre-treatment groups. I think that this would increase the power of your statistical analysis insofar as preventing any confounding results, allowing you to parse the post-treatment data more effectively. In addition, although 16S is sufficient, I think this may present some inconsistencies that hearken back to the Occam's razor concept. Perhaps utilizing a more thorough analysis of essential genes and/or the ITS region coupled with the 16S data would provide data with a greater level of accuracy, albeit this may be beyond the scope of the project.
Hi David,
I would recommend keeping your groups separate despite these data being pre-treatment groups. I think that this would increase the power of your statistical analysis insofar as preventing any confounding results, allowing you to parse the post-treatment data more effectively. In addition, although 16S is sufficient, I think this may present some inconsistencies that hearken back to the Occam's razor concept. Perhaps utilizing a more thorough analysis of essential genes and/or the ITS region coupled with the 16S data would provide data with a greater level of accuracy, albeit this may be beyond the scope of the project.
Why would you want to mess up your nice paired design? Be thankful your PI didn't get involved earlier and have you sample just the control mice pre-treatment, as they're all the same at this time-point.
Keep them separate.
I agree. Keep them separate. All the statistical analysis you can perform betweem your samples will be limited if you mix them. You should perform all the statistics in the separate samples. Maybe your PI have a single graphic in mind to compare the conditions and that's ok, but the deviation and variations of the sample should be there. You can use the principal component analysis in beta diversity analysis to see how different each sample is.
I agree that the mouse can be considered equivalent, but in the practice you'll realize that even then, they have differences and is statistically better to keep them separate.