Improving the Utility of Polygenic Risk Scores as a Biomarker for Alzheimer’s Disease

Abstract

The treatment of complex and multifactorial diseases constitutes a big challenge in day-to-day clinical practice. As many parameters influence clinical phenotypes, accurate diagnosis and prompt therapeutic management is often difficult. Significant research and investment focuses on state-of-the-art genomic and metagenomic analyses in the burgeoning field of Precision (or Personalized) Medicine with genome-wide-association-studies (GWAS) helping in this direction by linking patient genotypes at specific polymorphic sites (single-nucleotide polymorphisms, SNPs) to the specific phenotype. The generation of polygenic risk scores (PRSs) is a relatively novel statistical method that associates the collective genotypes at many of a person’s SNPs to a trait or disease. As GWAS sample sizes increase, PRSs may become a powerful tool for prevention, early diagnosis and treatment. However, the complexity and multidimensionality of genetic and environmental contributions to phenotypes continue to pose significant challenges for the clinical, broad-scale use of PRSs. To improve the value of PRS measures, we propose a novel pipeline which might better utilize GWAS results and improve the utility of PRS when applied to Alzheimer’s Disease (AD), as a paradigm of multifactorial disease with existing large GWAS datasets that have not yet achieved significant clinical impact. We propose a refined approach for the construction of AD PRS improved by (1), taking into consideration the genetic loci where the SNPs are located, (2) evaluating the post-translational impact of SNPs on coding and non-coding regions by focusing on overlap with open chromatin data and SNPs that are expression quantitative trait loci (QTLs), and (3) scoring and annotating the severity of the associated clinical phenotype into the PRS. Open chromatin and eQTL data need to be carefully selected based on tissue/cell type of origin (e.g., brain, excitatory neurons). Applying such filters to traditional PRS on GWAS studies of complex diseases like AD, can produce a set of SNPs weighted according to our algorithm and a more useful PRS. Our proposed methodology may pave the way for new applications of genomic machine and deep learning pipelines to GWAS datasets in an effort to identify novel clinically useful genetic biomarkers for complex diseases like AD.