My experiment was performed in 2 different farms and I have 16S rRNA sequences from these 2 farms. When I analysed the data, I observed strong farm effects. How to correct the farm effects to identify the true treatment effect? Is it OK to correct for the farm effect or need to analyse these 2 farms separately? I tried the combat function and got negative values in the feature table. I do not think this is OK. Can anyone help me with this?
Hi @Jeyamalar_Jeyanathan ,
Good question. This is a common issue in microbiome research (and beyond!). For sure things like block effects and farm effects are a common issue in agricultural research more generally, and there are statistical approaches to address this.
I would not "correct" for the farm effect, but include this as a covariate in your statistical tests where possible. E.g., when running a PERMANOVA test or ANOVA, you could include "farm" as a factor. This will account for variation between farms when testing for other effects.
That's right, your feature tables should only contain positive values. COMBAT was designed for batch correction of microarray data if I recall correctly, so is not really appropriate for microbiome data, which had different characteristics (e.g., compositionality).
For more general information, beyond my suggested solution above, I suggest looking into the literature on how to statistically handle block effects, as this is effectively what you have (though in this case they are entirely separate farms, not just blocks on the same farm).
Thank you for the suggestions Nicholas .