Clustering high dimensional mixed data to uncover sub-phenotypes:joint analysis of phenotypic and genotypic data

摘要

The LIPGENE-SU.VI.MAX study, like many others, recorded high dimensionalcontinuous phenotypic data and categorical genotypic data. LIPGENE-SU.VI.MAXfocuses on the need to account for both phenotypic and genetic factors whenstudying the metabolic syndrome (MetS), a complex disorder that can lead tohigher risk of type 2 diabetes and cardiovascular disease. Interest lies inclustering the LIPGENE-SU.VI.MAX participants into homogeneous groups orsub-phenotypes, by jointly considering their phenotypic and genotypic data, andin determining which variables are discriminatory. A novel latent variable model which elegantly accommodates high dimensional,mixed data is developed to cluster LIPGENE-SU.VI.MAX participants using aBayesian finite mixture model. A computationally efficient variable selectionalgorithm is incorporated, estimation is via a Gibbs sampling algorithm and anapproximate BIC-MCMC criterion is developed to select the optimal model. Two clusters or sub-phenotypes (`healthy' and `at risk') are uncovered. Asmall subset of variables is deemed discriminatory which notably includesphenotypic and genotypic variables, highlighting the need to jointly considerboth factors. Further, seven years after the LIPGENE-SU.VI.MAX data werecollected, participants underwent further analysis to diagnose presence orabsence of the MetS. The two uncovered sub-phenotypes strongly correspond tothe seven year follow up disease classification, highlighting the role ofphenotypic and genotypic factors in the MetS, and emphasising the potentialutility of the clustering approach in early screening. Additionally, theability of the proposed approach to define the uncertainty in sub-phenotypemembership at the participant level is synonymous with the concepts ofprecision medicine and nutrition.