An Approach to Analyze Longitudinal Zero-Inflated Microbiome Count Data Using Two-Stage Mixed Effects Models
- PDF / 928,247 Bytes
- 24 Pages / 439.37 x 666.142 pts Page_size
- 10 Downloads / 194 Views
An Approach to Analyze Longitudinal Zero‑Inflated Microbiome Count Data Using Two‑Stage Mixed Effects Models Jian Wang1 · Cielito C. Reyes‑Gibby2 · Sanjay Shete1,3,4 Received: 21 May 2019 / Revised: 7 August 2020 / Accepted: 16 September 2020 © International Chinese Statistical Association 2020
Abstract The developments of powerful next-generation technologies provide valuable resources for investigating the human microbiome. In particular, it has been of great interest to study the longitudinal changes in the microbiome and its association with risk factors and clinical outcomes (e.g., longitudinal oral microbiome abundance and oral mucositis). The challenges of such analysis include the zeroinflated microbial abundance counts data and the correlation among the longitudinal abundance counts collected across different time points within the same patient. The current approaches for longitudinal zero-inflated microbiome abundance data focused on testing the covariate-taxon associations (i.e., time-varying abundance as the dependent variable), but ignored the taxon-outcome associations (i.e., non-timevarying clinical outcome as the dependent variable). To address this question, we proposed a two-stage mixed effects model for analyzing zero-inflated longitudinal data and the clinical outcome of interest. In this model, the longitudinal microbial abundance count data are first modeled as a function of time using the zero-inflated negative binomial mixed effects model, and at the second stage, the summaries of the temporal patterns (e.g., random intercepts and slopes) are used in the regression models (e.g., linear, accelerate failure time) to assess their associations with the outcome. Simulations showed that the two-stage mixed effects model can provide accurate estimations for the regression coefficients of the association between the longitudinal trend of microbial abundance and the outcome. We applied the proposed approach to the study of longitudinal patterns in oral microbial abundance and oral mucositis in the patients with squamous cell carcinoma of the head and neck.
Electronic supplementary material The online version of this article (https://doi.org/10.1007/s1256 1-020-09295-y) contains supplementary material, which is available to authorized users. * Sanjay Shete [email protected] Extended author information available on the last page of the article
13
Vol.:(0123456789)
Statistics in Biosciences
Keywords Oral microbiome · Two-stage model · Zero-inflated negative binomial mixed effects model · Longitudinal count data
1 Introduction The human microbiome is associated with many complex disorders, such as obesity [1], cardiovascular disease [2], diabetes [3, 4], liver cirrhosis [5], inflammatory bowel disease [6], psoriasis [7], colorectal cancer [8, 9], as well as response to cancer immunotherapy [10–12]. Developments in next-generation sequencing technologies, such as 16S rRNA sequencing and shotgun sequencing, have allowed researchers to fully quantify microbiome composition using a cultur
Data Loading...
