| Sumario: | Purpose Predicting risk of Alzheimer's disease for the primary prevention and early intervention. Method We investigated the performance of combining multi-modal biomarkers obtained from GWAS study (n=14,000). We first predict the lifetime risk of AD using the ethnicity adjusted SNVs incorporating the prior genome-wide association (GWA) knowledge and predict the onset year for those with high lifetime risk by learning neuropsychological test scores and life-log (sleep quality and daily activity) information. Results and discussion When validated with three independent samples, the lifetime risk of AD using the proposed genomic prediction showed 75.3% of average accuracy (max 79.1%) with 74.8% and 73.7% of average sensitivity (max 80.5%) and specificity (73.7%), respectively. For those with high risk, we predicted onset within two years, our deep learning model showed 77.4% of accuracy. Strict blind test and application of the proposed model on GARD cohort shows its validity in application for screening and prevention of AD patients, especially in preclinical stage.
|
|---|