Hi everyone! Over the past year, the lab I work in has started to become more interested in mouse microbiome studies. The goal is to compare our disease models with wild type mice to see if there are differences.
So far my colleagues and I have analyzed data from a few studies and none have shown significant results, which I fear is the result of low sample sizes. One of our studies had two experimental groups with 3 mice each, another had 2 groups of 4, and another had 5 groups of 3-4 each.
Based on my search of the literature, it seems that many mouse microbiome studies have 5-6+ mice per group. And even when browsing this forum, it appears that a lot of researchers are using larger sample sizes than what my lab has done so far. Is this true? For those with experience in these types of studies, what sample sizes do you usually use? Is it reasonable to think that perhaps increasing sample size could lend different results? I know it’s possible that there truly aren’t any significant differences between these disease states and wild type, and that maybe adding a couple more mice to the study wouldn’t make any difference. I’m just trying to decide if it would be worth it to repeat these studies with more mice.
I’ve done some work on analyzing mouse studies. But, there are a few things to consider. First, your mice are coprophagic . (True story, I once walked in on my pet rat practicing coprophagy and he didn’t even know to look embarassed!) Practically, this means that cage becomes a potential unit for confounding your data. So, you need at least 2 cages per group to make sure the effect you’re seeing is not due to differences between cages. (Three per group would be better. Especially if your cages have variable numbers of mice.)
You will also have a statistical issue at smaller group sizes. The common rank-based tests don’t work well with fewer than 5 samples per group. So, if you go smaller, you’re going to violate those assumptions and they’re not going to behave. You have a similar issue in beta diversity; there’s a lower p-value limit that’s set by the number of samples. At 2 x 3, there are less than 1000 possible ways to shuffle the data, meaning that your p-value with 999 permutations doesn’t accurately reflect the distribution of the data…
Most of your differential abundance methods are also sensitive to small sample sizes, and you’re paying an FDR penalty every single time, so you’re trying to balance resolution with FDR and you rarely end up with a significant difference.
My best sample size recommendation is to think carefully about what you want to do and pick one or two experimental conditions (plus proper controls!) that let you address that problem. You want to be really careful about confounders (for example, all your mice need to be the same shipment if you’re getting them from somewhere else) and then you can address a sample size. At your current size, I would say that you’ve learned that you don’t have enough informationt continue. Personally, I try to avoid mouse studies with fewer than 10 animals and at least 2 cages per treatment condition because my experience has been that they’re just an exercise in frustration.
. (True story, I once walked in on my pet rat practicing coprophagy and he didn’t even know to look embarassed!) Practically, this means that cage becomes a potential unit for confounding your data. So, you need at least 2 cages per group to make sure the effect you’re seeing is not due to differences between cages. (Three per group would be better. Especially if your cages have variable numbers of mice.)