Genome wide association studies (GWAS) continue to reveal multiple human genomic loci and variants associated with the risk for Alzheimer's Disease (AD). However, identification of causal functional variants from GWAS remains an open problem since linkage disequilibrium (LD) between variants can lead to false positives. Most disease associated variants also lie in non-coding regions of the genome which are not as well characterized as protein coding genes. A major mechanism by which non-coding variants could influence AD predisposition is through the disruption of cis regulatory elements (CREs) which control gene expression. The activity of CREs can be highly specific to individual cell types, tissues, or cell states making it difficult to identify functional variants. The work in this thesis integrates GWAS data with cell type-specific epigenomics data from multiple cell types and develops sophisticated models to identify the cell types, variants, and CREs that influence predisposition to AD. Interestingly, the majority of AD associated variants predicted to have functional effects on CRE activity by the models have a greater predicted impact in peripheral immune cell types relative to brain resident cell types like microglia or neurons. Following up from this, this thesis also describes the development of methods to create an accurate biomarker for AD from CREs active in peripheral blood. Strikingly, incorporating CREs in regions genetically associated with AD into predictive models improves prediction accuracy of AD diagnosis significantly. We also apply transfer learning to improve our models by fine tuning models pretrained on open chromatin data on massively parallel reporter assay data. In addition to laying down best practices for model development in these areas, the work in this thesis suggests that peripheral immune cells themselves may mediate a component of AD predisposition and support their use as model systems to study the effects of AD associated variants.

194 pages

Thesis Committee:
Andreas Pfenning (Chair)
Ziv Bar-Joseph
Dennis Kostka (University of Pittsburgh)
Iliya Lefterov (University of Pittsburgh)

Russell Schwartz, Head, Computational Biology Department
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu