A Bayesian Mixture of Exponential Family Factor Models for Uncovering Disease Progression Subtypes.

Area of Science:

• Biostatistics
• Computational Biology
• Neuroscience

Background:

• Neurological disorders exhibit significant heterogeneity in biomarkers and clinical measures due to disease stage, individual characteristics, and latent subtypes.
• Understanding this heterogeneity is vital for accurate diagnosis and timely, targeted interventions.

Purpose of the Study:

• To develop a statistical model for integrating multi-modal data to characterize disease progression and identify patient subtypes in neurological disorders.
• To enable the learning of disease progression trajectories and temporal ordering of neurodegeneration.

Main Methods:

• Proposed a mixture exponential family trajectory model to integrate continuous (neuroimaging, microRNA sequencing), categorical (clinical symptoms), and ordinal (cognitive) markers.
• Utilized lower-dimensional latent factors and subtype-specific parameters within a mixture model framework.
• Employed a Bayesian estimation procedure with Markov chain Monte Carlo (MCMC) sampling for statistical inference.

Main Results:

• The model successfully characterizes patients within heterogeneous latent subgroups, reflecting distinct disease trajectories.
• Demonstrated the capability to describe nonlinear disease deterioration and provide temporal sequences of decline for various markers.
• Validated through extensive simulations and application to the Parkinson's Progression Markers Initiative (PPMI) dataset.

Conclusions:

• The proposed mixture model effectively captures complex heterogeneity in neurological disorders by integrating multi-modal data.
• This approach facilitates the identification of distinct patient subtypes and their specific disease progression pathways.
• Offers a robust framework for advancing early diagnosis and personalized treatment strategies in neurodegenerative diseases like Parkinson's disease (PD).