Premium
Genetic risk prediction of late‐onset Alzheimer’s disease based on tissue‐specific transcriptomic analysis and polygenic risk scores
Author(s) -
Jung SangHyuk,
Nho Kwangsik,
Kim Dokyoon,
Won HongHee
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.045184
Subject(s) - genome wide association study , disease , genetic association , computational biology , biology , single nucleotide polymorphism , gene , medicine , genetics , genotype
Background Large‐scale genome‐wide association studies (GWAS) have identified more than 25 genomic regions related to Alzheimer’s disease (AD). Recent studies showed polygenic risk score (PRS) could be used to identify individuals at high risk of AD. Despite this success, prediction and early intervention of AD still remain challenging. In this study, we constructed prediction models for AD with clinical features, PRS, and tissue‐specifically predicted gene expression levels, and evaluated prediction performance. Method We used whole‐genome sequencing data of 446 European participants (207 AD cases and 239 cognitively normal controls) from the Alzheimer’s disease Neuroimaging Initiative (ADNI), and performed quality control of genotyped data using PLINK (Table 1). PRS was derived based on a large‐scale GWAS summary statistics (Lambert et al. 2013) using PRS‐CS. We selected genes significantly associated with AD in brain using MetaXcan and the GWAS summary statistics. For these genes, we predicted gene expression levels in 13 brain regions for our study participants by using PrediXcan and GTEx V7. Then, we built a set of different prediction models that incorporated clinical features, PRS, and transcriptomic features using the Random Forest algorithm for the comparison. We evaluated prediction performance in 10‐fold cross‐validation. Result An AD prediction model using only demographic information and APOE ε4 status yielded an AUC of 0.726 (±0.055). As shown in Table 2, the inclusion of PRS significantly improved the prediction performance with an AUC of 0.758 (±0.0489). In addition, the inclusion of PRS and transcriptomic features produced much higher performance with an AUC of 0.794 (±0.061). Furthermore, the top 10 overlapping genes in 13 brain regions used in the prediction model were known AD susceptibility genes (Table 3). Conclusion Our study has improved the performance of existing AD genetic risk prediction models by incorporating tissue‐specific transcriptomic factors and PRS, suggesting that tissue‐specific transcriptomic factors and conventional PRS may be partly independent and complementary, so that both are important genetic risk predictors for the classification of AD. In particular, predicted brain expression factors could further provide an indication of tissue‐specific regulatory effects in AD.