In yukamoro/HTMMIOW: Hypothesis Test for Microbiome Mediation using Inverse Odds Weighting

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

This is a practical guide to to using HT-MMIOW: A Hypothesis Test approach for Microbiome Mediation using Inverse Odds Weighting.

Introduction

Mediation analyses quantifies the contribution of the microbiome in the "causal"" path between exposure and disease. HT-MMIOW is a hypothesis test that determines the statistical significance of the indirect effect of the microbiome. Briefly, this approach uses isometric log-ratio tranformation to account for the compositionality of microbiome data, and then uses UMAP to reduce its dimensionality. The resulting components serve as mediators in the mediation analysis. We use a mediation analysis technique that utilizes inverse odds ratio weighting to estimate the indirect effect of multiple mediators. We finally employ a permutation test to test the significance of the observed indirect effect.

Usage

Installation

You can install the development version of HTMMIOW from GitHub with:

# install.packages("devtools")
library(devtools)
install_github("yukamoro/HTMMIOW")

Application to simulated data

Let us apply HT-MMIOW to a simulated dataset. This dataset contains dichotomous exposure and dichotomous outcome variables for 300 subjects. The microbiome data contains counts (absolute abundance) for 400 taxa. Of the 400 taxa, 10 are true mediators.

library(HTMMIOW)
data("sim_data")

An exposure, outcome, and mediator variables are required to run the Perform_HTMMIOW function. Covariate data is optional. The user may specify the number of components to include in the mediation model and the number of permutations in the permutation test. Here we specify 2 components and 5000 permutations.

set.seed(10)
htmmiow_output = Perform_HTMMIOW(exposure = sim_data$sim_exposure,
                                 outcome = sim_data$sim_outcome,
                                 mediator = sim_data$sim_microbiome,
                                 confounder = NULL,
                                 umapcomponents = 2,
                                 B = 5000)

The outcome consists of three components:

1) tk: The estimated indirect effect, i.e. the observed test statistic for the permutation test 2) pk_2: The p-value of the permutation test 3) perm_null: Statistics estimated under the null hypothesis

In our simulated data example, the p-value is r htmmiow_output$pk_2, so we reject the null hypothesis and conclude that the microbiome is in fact a mediator in the causal path between exposure and outcome.

yukamoro/HTMMIOW documentation built on May 5, 2022, 6:47 a.m.