Genomic data analyses to understand genetic risks and protective factors of Alzheimer's disease

The overall objective of the proposed research is to use dataset NG00067 to conduct genetic research for better
understanding of genetic protective and risk factors underlying the Alzheimer’s disease (AD) continuum and use
artificial intelligence techniques to improve individual-level AD prediction with genomic features. Our central
hypothesis are 1) loss-of-function single nucleotide polymorphisms (SNP) of AD-risk genes, and their interactive
genes at the protein level, might confer protective effect against AD; and 2) deep convolutional neural network
(CNN) can deal with large and colinear feature space, thus the CNN classifier could potentially improve the
performance of individual-level AD continuum prediction with genetic features over non-deep classifiers.
To test these hypothesis, we plan to examine single loss-of-function mutation SNPs of AD-risk genes, and their
protein level interactors, as potential protective factors for AD (Aim1). More specifically, AD-associated genetic
risks/hazards will be obtained from previous literatures. Their protein-level interactive genes will be obtained from
the Reactome database. Loss-of-function SNPs associated with these genes will be further obtained from the
gnomAD database. Phenotypic comparisons along the AD continuum will be conducted among 0-copy, 1-copy,
and 2-copy mutation carriers of these SNPs.
In addition, we plan to develop a CNN-based deep learning technique to derive high-order meaningful genetic
features from the whole-genome or whole-exome sequencing data that could improve the individual-level disease
prediction (Aim2). For subjects with brain imaging or CSF biomarker data available from the NACC database, we
plan to further jointly evaluate the associations among genetic risks/protectors, brain imaging changes, and CSF
biomarkers along the AD continuum (Aim 3) to better understand the pathophysiology underlying AD onset and
progression. A separate data request has been made to the NACC database.

The purpose of this study is 1) to better understand genetic protective and risk factors underlying Alzheimer's
disease, and 2) to improve individual level disease predictions with artificial intelligence techniques. We seek to
identify AD protective genetic features through screening the disease phenotypes of loss-of-function mutation
carriers in subjects along the AD spectrum. We also would like to take advantages of the deep neural network in
capturing high-order meaningful features from large, high-dimensional, and co-linear feature spaces. Therefore,
we will utilize/adapt the well-established deep convolutional neural network classifier with genetic features to
predict individual-level disease status. These results will help us to better identify subjects at risks of AD, and
improve the accuracy and efficacy for personalized diagnosis, treatment, and prevention.