DirtyGenes: significance testing for gene or bacterial population composition changes using the Dirichlet distribution

High throughput sequencing, and quantitative polymerase chain reaction (qPCR), can detect changes in bacterial communities or the genes that they carry, between different environments or treatments. These methods are applied widely to microbiomes in humans, animals, soil and water; an important application is for changes in antimicrobial resistance genes (ARGs). However, at present, there is no statistical method to determine whether observed changes in the overall composition are significant, or result from random variations between samples. Therefore researchers are limited to graphical descriptions. We describe a novel statistical method to determine whether or not observed differences in bacterial populations or their genes are significant. It can be used with data from shotgun metagenomics, 16S characterisation or qPCR. It can also be used for experimental design, to calculate the number of samples needed in future experiments. We show its application to two example data sets. The first is published data on bacterial communities and ARGs in the environment, in which we show that there are significant changes in both ARG and community composition. The second is a new data set on seasonality in bacterial communities and ARGs in hooves from four sheep. While the observed differences are not significant, we show that a minimum group size of eight sheep in a future experiment would provide sufficient power to observe significant changes, should the already observed changes be true. This method has broad uses for statistical testing and experimental design in experiments on changing microbiomes, including for studies on antimicrobial resistance.