Alzheimer's disease(AD) subtype analysis using genome sequencing data

Objectives of the Proposed Research
Alzheimer’s disease (AD) is a common degenerative disease, causing irreversible dementia. Early diagnosis is difficult due to a long asymptomatic period and requires invasive, expensive procedures. A screening method to classify high-risk groups for early AD diagnosis is needed.
Study Design
Early AD risk prediction can use genomic variants like the Polygenic Risk Score (PRS), which predicts high-risk groups but shows performance differences due to genetic heterogeneity and ethnic specificity. To address this, ethnicity-specific analysis is considered and validated with different ethnic datasets.
This study aims to develop Korea-specific PRS models for early AD risk prediction using genomic data from a Korean cohort and the ADSP. Trans-ethnic genomic data will be created by combining GARD and ADSP data, including African American (AA), non-Hispanic Whites (NHW), and East Asian (EA) data. Cross-validation (CV)
analysis will divide data into training and test sets. Genomic variants' importance (e.g., p-values, BLUP) will be calculated, and selected variants applied to PRS. PRS models will be evaluated using CV-divided test data to select the best model. Trans-ethnic and ethnicity-specific PRS models will be validated using reserved validation data.
Analysis Plan
The proposal aims to identify ethnicity differences in genomic prediction built with Caucasian-centric GWA SNVs and improve the model for trans-ethnic groups, particularly East Asians. A Bayesian machine learning approach transfers genetic risk model knowledge from the NHW dataset to other ethnic groups for better accuracy. Genotype datasets from all ancestry groups are used together. Instead of trans-ethnic meta-analysis, the approach by Gim et al. is adopted. Each ethnic group dataset is divided for cross-validation. Training datasets are analyzed to evaluate p-values and BLUP of SNVs. Summary statistics are used to build the prediction model and apply nested-CV for model selection. The best model for each ethnic group is tested using the test dataset. Data is analyzed similarly by learning from ethnic-specific variants and building a prediction model with the new method.

Alzheimer’s disease (AD) is the leading cause of dementia and is irreversible once symptoms appear. A long asymptomatic period of AD complicates early diagnosis requiring invasive and costly procedures like CSF extraction or PET scans. Therefore, a screening method to identify high-risk groups for early AD diagnosis is necessary.

One approach uses the Polygenic Risk Score (PRS), which calculation is based on multiple genomic variants associated with AD. However, PRS predictions vary significantly (60-80%) due to genetic heterogeneity and ethnic specificity. Thus, data from multiple ethnicities must be analyzed. Although Asia accounts for over 50% of global dementia cases, most large-scale AD cohorts are predominantly White, lacking studies on Asians.

This study aims to develop trans-ethnic and ethnicity-specific PRS models for early AD risk prediction using genomic data from the GARD cohort, centered on Koreans, and the ADSP, which includes various European ethnicities. It investigates AD’s genetic heterogeneity due to ethnic differences and proposes methods to adjust for variability.