Computational CRISPR deep learning simulation of the likelihood of Alzheimer’s disease using chromosome 19 GWAS

Abstract

AbstractBackgroundSeveral genetic variants in chromosome 19 have been determined as independent Alzheimer’s disease (AD) risk factors. It is presumed that different combinations amongst them would further impact AD risk and progression at the individual level. In order to advance AD precision medicine, it is critical to understand the underlying complex genetic mechanisms. Deep learning technology is capable of not only identifying such variants but also defining complicated interactions that could further influence disease progression.MethodWe applied a novel deep learning model to the 266,161 SNP chromosome 19 data of 313 AD and 457 cognitively unimpaired (CU) ADNI patients. By using occlusion mapping we identified the SNPs that contributed to predictive accuracy of AD vs. CU and sequentially replaced each of them with all other biologically plausible alleles to most precisely measure the impact of the variability at the SNP level. Next, we examined the possible epistatic interactions on the SNP level on AD likelihood by co‐occlusion of pairs of SNPs and investigating the additional changes on predictive accuracy.ResultRs56131196 (APOC1) emerged as the most powerful chromosome 19 AD SNP. 79.69% (51/64) of the predicted AD patients were heterozygous, i.e., AG, at this locus. In contrast, 88.89% (80/90) of the predicted CU patients had GG genotype at rs56131196. Substituting AG for GG in predicted AD participants resulted in changing the diagnostic prediction to CU in 36% of the cases. Substituting GG for AA in predicted CU participants resulted in changing the diagnostic prediction to AD in 6% of the cases. The model further identified that AG→GG replacement further affects the likelihood of AD by interactions with SNPs in other genes including APOC1/APOC1P1, TOMM40, and ERCC1/CD3EAP.ConclusionWe tested the APOC1 rs56131196 genotype risk for AD by creating a simulator analogous to CRISPR. This innovative approach could provide a potential in silico CRISPR predictive modeling and help advance personalized preventive gene therapy for AD.