Microbiome differential abundance analysis by zero-inflated quantile rank-score based test

Studying the role of microorganisms in the development of diseases such as obesity in human populations is at an all-time high. The very objective of statistical analysis on microbial data is to identify differentially abundant taxa among certain clinical conditions, to guide follow-up pathway analysis. However, the microbiome differential abundance analysis (MDA) is challenging. First, microbiome data needs to be normalized because of differences in read depths. There are many different resampling or scaling normalization methods, and the performance of existing MDA approaches are highly dependent on the normalization choices, which impedes the comparison among various studies. The introduction of a robust method to the normalization is then of interest. Second, microbiome data is complex, usually zero-inflated, dispersed and high-dimensional. It is hopeless to determine a one-size-fit-all parametric method for all taxa; also, these mean-based methods are insufficient to detect the heterogeneous association between the clinical condition and microbial abundance. Quantile regression is a powerful alternative to deal with heterogeneity, as it does not require likelihood specification, and one can use it to examine various locations of the abundance distribution. In this paper, we propose to use a quantile rank-score based test (ZIQRank) under a two-part quantile regression model to analyze the microbiome data processed by any normalization method. The tool consists of a valid test in logistic regression for the zero-inflation, and a series of rank-score based tests on multiple quantiles of the positive part with adjustment for zero-inflation. We applied ZIQRank to study the association between gut microbiota and high blood pressure, confirmed that it increases the power with well-controlled Type I error and is robust to the normalization method. It identified most of the taxa detected by the existing MDA methods and complemented them with finding additional taxa on which hypertension has heterogeneous effects.