GCCN: Graph Capsule Convolutional Network for Progressive Mild Cognitive Impairment Prediction and Pathogenesis...

Area of Science:

• Neuroscience
• Computational Biology
• Artificial Intelligence

Background:

• Mild cognitive impairment (MCI) is a precursor to dementia, necessitating accurate prediction and understanding of its pathogenesis.
• Identifying key pathogenic factors and their interactions is crucial for developing effective interventions.
• Current methods may not fully capture the complex, heterogeneous nature of disease progression.

Purpose of the Study:

• To propose a novel Graph Capsule Convolutional Network (GCCN) for predicting MCI to dementia progression.
• To identify the underlying pathogenesis and key risk factors involved in disease development.
• To leverage heterogeneous pathogenic information for a more comprehensive disease model.

Main Methods:

• Constructed heterogeneous pathogenic information association graphs using risk genes and brain regions as nodes.
• Developed graph capsules by projecting information into disentangled latent components representing format and intensity.
• Employed GCCN to model information flow among pathogenic factors and identify disease-driving pathways via dynamic routing.

Main Results:

• GCCN demonstrated significant advancements over existing methods on public datasets.
• The identified pathogenic factors were evidential and strongly correlated with progressive MCI.
• The dynamic routing mechanism successfully captured discriminative pathogenic information flows.

Conclusions:

• GCCN offers a powerful new approach for predicting MCI to dementia progression.
• The method effectively identifies key pathogenic factors and elucidates disease mechanisms.
• This work provides a foundation for developing targeted diagnostic and therapeutic strategies.