Multilayer Exponential Family Factor models for integrative analysis and learning disease progression.

Area of Science:

• Neuroscience
• Biostatistics
• Computational Biology

Background:

• Neurological disorder diagnosis often relies on late-stage symptoms, hindering early intervention.
• Biomarkers and subtle clinical changes may indicate disease onset earlier.
• Integrating multidomain markers is challenging but vital for understanding early disease progression.

Purpose of the Study:

• To develop a method for leveraging multidomain markers to learn early neurological disease progression.
• To address the challenge of integrating heterogeneous data types for disease phenotyping.
• To enable early detection and intervention by understanding premanifest disease stages.

Main Methods:

• Proposed a hierarchical Multilayer Exponential Family Factor (MEFF) model.
• Integrated heterogeneous measures (clinical, cognitive, neuroimaging, blood biomarkers).
• Employed approximate inference techniques for fitting the model to large-scale data.

Main Results:

• The MEFF model successfully integrated diverse data types for Parkinson's disease (PD).
• Identified shared and domain-specific latent factors for robust phenotyping.
• Learned lower-dimensional representations and the temporal ordering of neurodegeneration in PD.

Conclusions:

• The developed MEFF model offers a robust approach for early neurological disease progression analysis.
• Facilitates clinically meaningful patient subgrouping and understanding of disease trajectories.
• Provides a foundation for developing targeted interventions at premanifest stages of neurological disorders.