Multi-omics subtyping pipeline for chronic obstructive pulmonary disease.

Area of Science:

• Pulmonary Medicine
• Genomics
• Systems Biology

Background:

• Chronic Obstructive Pulmonary Disease (COPD) is a leading cause of mortality with complex clinical heterogeneity.
• Existing clinical classifications do not fully capture the molecular underpinnings of COPD phenotypes.

Purpose of the Study:

• To investigate whether multi-omics data (transcriptomics, proteomics, metabolomics) can define molecular subtypes of COPD with distinct clinical characteristics.
• To explore the utility of integrating different omics datasets for a deeper understanding of COPD heterogeneity.

Main Methods:

• Utilized transcriptomics, proteomics, and metabolomics data from 3,278 COPDGene cohort subjects.
• Applied autoencoder embeddings and subspace clustering (MineClus) to omics datasets, individually and combined.
• Employed recursive feature selection and Support Vector Machines to identify significant molecular features and clinical associations.

Main Results:

• Single omics clustering typically yielded two distinct clusters, each defined by unique molecular pathways (e.g., sphingomyelin in metabolomics, immune/bacterial responses in transcriptomics).
• Integrating omics data at a later stage identified COPD subtypes associated with variations in age, disease severity, lung function (DLCO), and atrial fibrillation.
• Early integration of omics data did not reveal clinically distinct subtypes.

Conclusions:

• Multi-omics data can define molecularly distinct COPD subtypes with varying clinical phenotypes.
• The integration of transcriptomics, proteomics, and metabolomics provides valuable insights into the molecular basis of COPD heterogeneity.
• This approach enhances our understanding of COPD beyond traditional clinical classifications.