Identification of Associations Between Peripheral Blood Gene Expression and Cerebrospinal Fluid Biomarkers in Alzheimer's Disease Using an Improved Joint Multi-Task Sparse Canonical Correlation Analysis Algorithm.

Abstract

Alzheimer's disease (AD) is an irreversible neurodegenerative disorder, and early diagnosis is crucial for effective clinical intervention. Traditional diagnostic methods involve detecting living brain tissue across the blood-brain barrier, but these invasive procedures cause unavoidable damage to patients. Genetic biomarkers in peripheral blood may provide valuable insights into brain lesions, potentially offering a non-invasive method for early AD diagnosis. The aim of this study is to propose an improved joint multi-task sparse canonical correlation analysis (MTSCCA) algorithm to identify significant genetic biomarkers in peripheral blood that correlate with brain markers of AD, such as cerebrospinal fluid (CSF) markers. This approach aims to accurately predict AD and assess disease progression. The study employs a multi-task sparse canonical correlation analysis (MTSCCA) approach with separate analyses for AD and healthy controls. Both tasks are constrained with class-consistent and class-specific conditions to identify significant features for each diagnostic group. To enhance robustness, the Laplacian matrix constraints were incorporated into the MTSCCA-LR algorithm to reduce noise in genetic data. The proposed algorithm identifies key differentially expressed genes (DEGs) that are involved in pathways closely linked to AD pathogenesis. These genes have specific diagnostic significance. Validation of these genes for predicting CSF markers was conducted using two regression models, showing good predictive accuracy. Furthermore, a Support Vector Machine (SVM) classifier was used to classify the two diagnostic groups, demonstrating high classification accuracy. The Top 20 genes identified using the proposed algorithm were used to construct an AD diagnostic model, which exhibited strong potential for non-invasive AD diagnosis, with significant implications for clinical practice. The code and example data of the proposed algorithm have been made publicly available on GitHub ( https://github.com/Zoe491/Improved-MTSCCA1 ).