A Machine Learning Method to Identify Genetic Variants Potentially Associated With Alzheimer's Disease

Bradley Monk; Andrei Rajkovic; Semar Petrus; Aleks Rajkovic; Terry Gaasterland; Roberto Malinow

doi:10.3389/fgene.2021.647436

A Machine Learning Method to Identify Genetic Variants Potentially Associated With Alzheimer's Disease

Front Genet. 2021 Jun 14:12:647436. doi: 10.3389/fgene.2021.647436. eCollection 2021.

Authors

Bradley Monk^{1

2}, Andrei Rajkovic³, Semar Petrus⁴, Aleks Rajkovic⁵, Terry Gaasterland⁴, Roberto Malinow^{1

6}

Affiliations

¹ Department of Neurosciences, Center for Neural Circuits and Behavior, School of Medicine, University of California, San Diego, San Diego, CA, United States.
² Cognitive Science & Psychology IDP, University of California, San Diego, San Diego, CA, United States.
³ Department of Computer Science, Royal Holloway, University of London, Egham, United Kingdom.
⁴ Institute for Genomic Medicine, Scripps Institution of Oceanography, University of California, San Diego, San Diego, CA, United States.
⁵ Department of Pathology, Department of Obstetrics, Gynecology and Reproductive Sciences, University of California, San Francisco, San Francisco, CA, United States.
⁶ Section of Neurobiology, Division of Biological Sciences, University of California, San Diego, San Diego, CA, United States.

Abstract

There is hope that genomic information will assist prediction, treatment, and understanding of Alzheimer's disease (AD). Here, using exome data from ∼10,000 individuals, we explore machine learning neural network (NN) methods to estimate the impact of SNPs (i.e., genetic variants) on AD risk. We develop an NN-based method (netSNP) that identifies hundreds of novel potentially protective or at-risk AD-associated SNPs (along with an effect measure); the majority with frequency under 0.01. For case individuals, the number of "protective" (or "at-risk") netSNP-identified SNPs in their genome correlates positively (or inversely) with their age of AD diagnosis and inversely (or positively) with autopsy neuropathology. The effect measure increases correlations. Simulations suggest our results are not due to genetic linkage, overfitting, or bias introduced by netSNP. These findings suggest that netSNP can identify SNPs associated with AD pathophysiology that may assist with the diagnosis and mechanistic understanding of the disease.

Keywords: Alzheimer’s; disease; machine learning; neural network; polygenic.

Grants and funding

I 904/FWF_/Austrian Science Fund FWF/Austria