Model-based multifacet clustering with high-dimensional omics applications.

Area of Science:

• Computational biology
• Bioinformatics
• Statistical genetics

Background:

• High-dimensional omics data present complex structures, often leading to multiple valid sample groupings based on different feature subsets.
• Traditional clustering methods generate a single solution, failing to capture the multifaceted nature of biological data.

Purpose of the Study:

• To develop a novel model-based multifacet clustering (MFClust) method for high-dimensional omics data.
• To address the limitations of conventional clustering in identifying multiple, simultaneous cluster structures.

Main Methods:

• Proposed MFClust method utilizes a mixture of Gaussian mixture models.
• The first mixture component assigns features to facets, while the second assigns samples to clusters within facets.
• Validated through simulation studies and applied to transcriptomic datasets.

Main Results:

• MFClust demonstrated superior accuracy in both facet and cluster assignments compared to conventional methods in simulations.
• Application to postmortem brain and lung disease transcriptomic data revealed clinically relevant multifacet clustering structures.
• Identified novel biological insights and potential hypotheses for further research.

Conclusions:

• MFClust effectively captures complex, multifacet clustering patterns in high-dimensional omics data.
• The method enhances biological discovery by revealing hidden structures associated with clinical variables.
• Offers a powerful tool for analyzing complex biological datasets in disease research.