Microbial Risk Score for Capturing Microbial Characteristics, Integrating Multi-omics Data, and Predicting Disease Risk

Abstract

BackgroundWith the rapid accumulation of microbiome-wide association studies, a great amount of microbiome data are available to study the microbiomes role in human disease and advance the microbiomes potential use for disease prediction. However, the unique features of microbiome data hinder its utility for disease prediction. MethodsMotivated from the polygenic risk score framework, we propose a microbial risk score (MRS) framework to aggregate the complicated microbial profile into a summarized risk score that can be used to measure and predict disease susceptibility. Specifically, the MRS algorithm involves two steps: 1) identifying a sub-community consisting of the signature microbial taxa associated with disease, and 2) integrating the identified microbial taxa into a continuous score. The first step is carried out using the existing sophisticated microbial association tests and pruning and thresholding method in the discovery samples. The second step constructs a community-based MRS by calculating alpha diversity on the identified sub-community in the validation samples. Moreover, we propose a multi-omics data integration method by jointly modeling the proposed MRS and other risk scores constructed from other omics data in disease prediction. ResultsThrough three comprehensive real data analyses using the NYU Langone Health COVID-19 cohort, the gut microbiome health index (GMHI) multi-study cohort, and a large type 1 diabetes cohort separately, we exhibit and evaluate the utility of the proposed MRS framework for disease prediction and multi-omics data integration. In addition, the disease-specific MRSs for colorectal adenoma, colorectal cancer, Crohns disease, and rheumatoid arthritis based on the relative abundances of 5, 6, 12, and 6 microbial taxa respectively are created and validated using the GMHI multi-study cohort. Especially, Crohns disease MRS achieves AUCs of 0.88 ([0.85-0.91]) and 0.86 ([0.78-0.95]) in the discovery and validation cohorts, respectively. ConclusionsThe proposed MRS framework sheds light on the utility of the microbiome data for disease prediction and multi-omics integration, and provides great potential in understanding the microbiomes role in disease diagnosis and prognosis.