Multi-omics subtyping pipeline for chronic obstructive pulmonary disease

• DOI: 10.1371/journal.pone.0255337
• Publication Date: 2021-08-25T17:26:28Z
• Authors: Lucas A. Gillenwater; Shahab Helmi; Evan Stene; Katherine A. Pratte; Yonghua Zhuang; Ronald P. Schuyler; Leslie Lange; Peter J. Castaldi; Craig P. Hersh; Farnoush Banaei-K...; Russell P. Bowler; Katerina J. Kechris
• Keywords: Omics; Metabolome; Biomarker Discovery; Subtyping

Chronic Obstructive Pulmonary Disease (COPD) is the third leading cause of mortality in United States; however, COPD has heterogeneous clinical phenotypes. This first large scale attempt which uses transcriptomics, proteomics, and metabolomics (multi-omics) to determine whether there are molecularly defined clusters with distinct phenotypes that may underlie heterogeneity. Subjects included 3,278 subjects from COPDGene cohort at least one following profiles: whole blood transcriptomes (2,650 subjects); plasma proteomes (1,013 metabolomes (1,136 subjects). 489 had all three contemporaneous -omics profiles. Autoencoder embeddings were performed individually for each dataset. Embeddings underwent subspace clustering using MineClus, either by or combined, followed recursive feature selection based on Support Vector Machines. Clusters tested associations variables. Optimal single typically resulted two clusters. Although was overlap individual cluster membership, tended be unique molecular pathways. For example, prominent features metabolome-based sphingomyelin, while key transcriptome-based related immune bacterial responses. We also found when we integrated data a later stage, identified subtypes varied age, severity disease, addition diffusing capacity lungs carbon monoxide, precent atrial fibrillation. In contrast, an earlier stage treating sets equally, no differences between subtypes. Similar clustering, revealed multiple heterogenous phenotypes, show tend define patients different characteristics. Thus, integrating these affords additional insight into nature its