A robust and flexible nonparametric approach for longitudinal microbiome data analysis

Md.  M. Hosain${}^a$, H. Koh${}^b$, and T.  Park${}^a$

${}^a$Department of Statistics, Seoul National University, Seoul, Korea, ${}^b$Department of Applied Mathematics and Statistics, The State University of New York, Korea (SUNY Korea), Incheon, Korea

Recent improvements and affordable sequencing technology have made it possible to conduct metagenomic studies on associations between members of microbial community with human health or disease in depth.

Longitudinal studies are essential to understand the dynamic relationship between microbiota and host phenotypes over time. However, most existing methods suffer from the inherent complexity, sparsity, over-dispersion, and high-dimensionality of longitudinal microbiome data, and therefore tend to address some but not all critical data characteristics. In this study, we propose the microbiome residual permutation and power transformation (MiRP) framework and two omnibus tests, including MiRP-O and MiRP-WO for association testing in longitudinal microbiome studies. The proposed methods employ generalized estimating equations (GEE), residual permutations, and power transformation of the microbial taxa to account for covariate effects and longitudinal correlation. MiRP-O uses the minimum p-value across various power-transformed abundances, whereas MiRP-W-O employs weighted averaging to enhance robustness.

We applied our proposed methods, MiRP-O and MiRP-W-O, to a simulation study and real data analyses. Through simulation studies, we showed that both MiRP-O and MiRP-W-O well control the Type I error rates and have higher power compared to other existing methods. Real data applications showed that methods were able to capture important microbial taxa related to phenotypes, indicating the effectiveness of the methods. MiRP-O and MiRP-W-O are both robust and powerful for capturing phenotype-related microbial markers in longitudinal microbiome studies.

Keywords: Microbiome, Longitudinal data, Relative abundance, Differential abundance analysis, Permutation test.