|
Abstract:
|
Microbiome data that characterize human (or animal) health and disease have been gaining enormous popularity among scientists. While novel statistical research is active in this field, easy access is a method from such as differential expression in genomics. The fundamental difference, however, is that the differential abundance in microbiome means the difference in relative abundances based on taxon counts, needing special attention to the “compositionality.” With such data, controlling false discovery rate has proven dif- ficult, we aim to examine the well-known Benjamini and Hochberg (1995) procedure under high dimensional compositional responses. First, compositionality makes normal theory test inappropriate. Second, compositionality creates negative correlations, questioning the independence of simultaneous hypotheses. Third, consider block design. Dependent samples are common in practice while most examples in literature are independent sample tests. Fourth, check subcompositionality if the testing results are robust to different subsets of variables, differently reported by labs. Overall, the coveats for the multiple testing in microbiome data are addressed.
|
