There is increasing interest in investigating how the compositions of microbial communities are associated with human health and disease. Although existing methods have identified many associations, a proper choice of a phylogenetic distance is critical for the power of these methods. To assess an overall association between the composition of a microbial community and an outcome of interest, we present a novel multivariate testing method called aMiSPU, that is joint and highly adaptive over all observed taxa and thus high powered across various scenarios, alleviating the issue with the choice of a phylogenetic distance. Our simulations and real-data analyses demonstrated that the aMiSPU test was often more powerful than several competing methods while correctly controlling type I error rates. The R package MiSPU is available at https://github.com/ChongWu-Biostat/MiSPU and CRAN.
Bibliographical noteFunding Information:
The authors thank Ni Zhao for helpful discussions on MiRKAT and sharing the R code implementing MiRKAT and simulations. This research was supported by National Institutes of Health grants R01GM113250, R01HL105397, and R01HL116720, by the Center for Individualized Medicine at Mayo Clinic, and the Minnesota Supercomputing Institute.