ci-fGBD: Cluster-Integrated Fast Generalized Bruhat Decomposition for Multimodal Data Clustering in Alzheimer's...

Area of Science:

• Biomedical data science
• Computational biology
• Neuroscience

Background:

• Multimodal biomedical datasets, common in neurodegenerative diseases, pose challenges for patient stratification due to missing data, high dimensionality, and biases.
• Existing clustering methods often require extensive preprocessing and struggle to integrate heterogeneous data types effectively.

Purpose of the Study:

• To introduce ci-fGBD (Cluster-Integrated Fast Generalized Bruhat Decomposition), a novel framework for stratifying heterogeneous patient populations using multimodal data.
• To develop a method that natively handles block-structured, multimodal datasets and harmonizes contributions across diverse data types.

Main Methods:

• ci-fGBD is a matrix factorization and clustering framework extending the classical Bruhat decomposition.
• It jointly learns latent representations and patient clusters, automatically harmonizing contributions from neuroimaging, cognitive assessments, genomics, wearable sensors, and environmental exposures.

Main Results:

• Benchmarking on real-world datasets demonstrates ci-fGBD's superior performance compared to standard methods.
• The framework consistently identifies clinically meaningful subgroups in Alzheimer disease cohorts.
• ci-fGBD captures subtle biological, cognitive, and demographic heterogeneity with enhanced interpretability and robustness.

Conclusions:

• ci-fGBD offers a robust and interpretable solution for stratifying complex patient populations using multimodal biomedical data.
• The framework effectively addresses challenges posed by missing values, high dimensionality, and modality-specific biases in neurodegenerative disease research.