HIP: a method for high-dimensional multi-view data integration and prediction accounting for subgroup heterogeneity

Area of Science:

• Bioinformatics
• Statistical genetics
• Computational biology

Background:

• Complex diseases exhibit subgroup disparities (e.g., sex, race) affecting disease course and outcomes.
• Current integrative analysis methods often overlook subgroup heterogeneity and fail to model associations between different data views (e.g., genomics, proteomics).

Purpose of the Study:

• To develop and apply a statistical approach, Heterogeneity in Integration and Prediction (HIP), for joint association and prediction in multi-view data.
• To identify molecular signatures (proteins, genes) shared by or specific to subgroups, accounting for heterogeneity.
• To investigate sex-specific molecular mechanisms contributing to chronic obstructive pulmonary disease (COPD).

Main Methods:

• Proposed HIP, a statistical method for integrative analysis of multi-view data.
• Leveraged strengths from different data views to identify subgroup-specific and shared molecular signatures.
• Applied HIP to proteomics and gene expression data in COPD, considering sex as a subgroup variable and airway wall thickness as the outcome.

Main Results:

• Identified proteins and genes that are common across and specific to males and females in COPD.
• Discovered molecular signatures implicated in COPD and potential new insights into sex-based mechanisms of the disease.
• Demonstrated HIP's capability to account for subgroup heterogeneity, rank variable importance, handle continuous outcomes, and adjust for covariates.

Conclusions:

• HIP effectively addresses subgroup heterogeneity in multi-view data analysis.
• The method enhances the identification of molecular signatures relevant to specific subgroups and disease mechanisms.
• HIP offers a powerful tool for multiomics research, particularly in understanding health disparities and complex diseases like COPD.