A fast integrative clustering and feature selection approach for high-dimensional multiview data.

Area of Science:

• Biomedical Data Science
• Computational Biology
• Statistical Genetics

Background:

• Cluster analysis is crucial for identifying disease subtypes in heterogeneous biomedical data.
• High dimensionality, multimodality, and computational complexity present challenges for current clustering methods.
• Advanced data science necessitates novel approaches for integrating diverse biological datasets.

Purpose of the Study:

• To propose a fast integrative clustering method, iClusterVB, using variational Bayesian inference.
• To enable feature selection in high-dimensional settings with mixed data types (continuous, categorical, count).
• To demonstrate the utility of iClusterVB in identifying clinically relevant cancer subtypes and biomarkers.

Main Methods:

• Developed iClusterVB, an integrative clustering approach based on variational Bayesian inference.
• Integrated multiple datasets for clustering and performed feature selection.
• Utilized mixed data types including continuous, categorical, and count data.

Main Results:

• iClusterVB demonstrated advantages over six competing methods in simulation studies.
• Applied to three real-life studies, identifying key features and cancer subtypes.
• Showcased association between identified subtypes and distinct survival probabilities.

Conclusions:

• iClusterVB offers a computationally efficient and effective solution for integrative clustering in high-dimensional biomedical data.
• The method facilitates the discovery of novel disease subtypes and predictive biomarkers.
• A user-friendly R package and tutorial are available for practical implementation.