|
Abstract:
|
Cluster analysis of omics data is an initial step to identify novel disease subtypes towards precision medicine. The classical unsupervised clustering strategy concerns primarily the identification of subpopulations that have similar patterns in selected genes. However, as the genes correspond to confounders (e.g. gender, age) may dominate clustering, the resulting clusters may or may not capture clinically meaningful disease subtypes. This gives rise to a fundamental problem: can we find a subtyping procedure guided by a specific disease outcome? We propose a unified latent generative model to perform outcome-guided disease subtyping constructed from omics data, which guarantees the resulting subtypes to concern the disease of interest. The hierarchical generative model with mixture regression has embedded regularization for gene selection. A modified EM algorithm is applied for numerical computation and parameter estimation. Extensive simulations and application to a complex lung disease dataset with transcriptomic and phenomic data demonstrate advantages of OG-Clust to identify clinically relevant disease subtypes suitable to explore toward precision medicine.
|
