A Novel Longitudinal Phenotype-Genotype Association Study Based on Deep Feature Extraction and Hypergraph Models for...

Area of Science:

• Neuroscience
• Genetics
• Biomedical Engineering

Background:

• Traditional Alzheimer's disease (AD) research often uses linear models, neglecting dynamic brain changes over time.
• Existing methods fail to capture complex, high-order correlations in longitudinal brain imaging and genetic data.

Purpose of the Study:

• To develop a novel method, Deep Subspace reconstruction with Hypergraph-Based Temporally-constrained Group Sparse Canonical Correlation Analysis (DS-HBTGSCCA), for analyzing longitudinal AD data.
• To uncover deep associations between time-varying brain phenotypes and genotypes in Alzheimer's disease.
• To identify novel AD biomarkers by leveraging dynamic, high-order correlations.

Main Methods:

• Applied deep subspace reconstruction to capture nonlinear properties of brain imaging and genetic data.
• Utilized hypergraphs to mine high-order correlations within the reconstructed data.
• Incorporated temporal constraints for analyzing longitudinal data, focusing on dynamic changes.

Main Results:

• The DS-HBTGSCCA method successfully extracted valuable time-series correlations from AD neuroimaging data.
• Identified potential AD biomarkers across multiple time points, demonstrating the algorithm's efficacy.
• Regression analysis confirmed strong relationships between identified brain regions and genes, validating the approach.

Conclusions:

• The proposed DS-HBTGSCCA method effectively analyzes dynamic, high-order correlations in longitudinal AD data.
• Deep subspace reconstruction enhances the discovery of nonlinear relationships and improves biomarker identification.
• This approach offers a powerful tool for advancing Alzheimer's disease research and biomarker discovery.