Integrative Generalized Convex Clustering Optimization and Feature Selection for Mixed Multi-View Data.

Area of Science:

• Computational Biology
• Data Science
• Machine Learning

Background:

• Mixed multi-view data involves multiple feature sets from the same samples.
• Integrative clustering aims to find common structures across these diverse data sources.
• Existing methods may not fully leverage the richness of multi-view data or handle high dimensionality effectively.

Purpose of the Study:

• To develop a novel convex clustering framework for integrating mixed multi-view data.
• To introduce a feature selection mechanism for high-dimensional multi-view datasets.
• To improve the discovery of common group structures within samples.

Main Methods:

• Propose Integrative Generalized Convex Clustering Optimization (iGecco) using view-specific losses and a joint convex fusion penalty.
• Develop iGecco+ incorporating an adaptive shifted group-lasso penalty for feature selection.
• Utilize a generalized multi-block ADMM algorithm for efficient model fitting on large datasets.

Main Results:

• iGecco effectively integrates diverse data views to identify common sample groupings.
• iGecco+ demonstrates superior performance in feature selection and integrative clustering for high-dimensional data.
• The proposed ADMM algorithm provides efficient solutions for large-scale multi-view clustering.

Conclusions:

• iGecco and iGecco+ offer a robust and mathematically grounded approach to integrative clustering.
• Feature selection via iGecco+ is crucial for improving clustering performance in high-dimensional multi-view settings.
• The methods show promise in applications like text mining and genomics.