B1683
Title: Predicting risk groups for time to event data using microbiome biomarkers: Methodology and software development
Authors: Thi Huyen Nguyen - Hasselt university (Belgium) [presenting]
Abstract: Identifying taxa that can be used to predict the time to develop Type 1 Diabetes (T1D) using various forms of microbiome data has been widely discussed in the literature, but not fully developed. The aim of the analysis is to clarify whether individuals at high or low risk of developing T1D in a follow-up control experiment in which subjects are randomized into two treatment groups and the time to develop T1D is monitored. Several methods are presented to estimate the microbiome risk score for the time to develop T1D such as the majority voting technique, LASSO, elastic net, supervised principle component analysis (SPCA), and supervised partial least squares analysis (SPLS). All estimation methods were evaluated within a l-fold Monte Carlo cross-validation (MCCV) loop. Within the evaluation process, validation is accessed using the hazard ratios (HR) distribution of the test set. The resampling-based inference is implemented using the permutation technique. A software tool to conduct such an analysis is not available yet. The R package MicrobiomeSurv is a user-friendly data analysis tool, both for modelling and visualization for this type of application. The package can be used for analysis at any level of interest in the microbiome ecosystem (OTUs, family, kingdom level, etc.).
