Submission
| Title: | Using an end-to-end deep learning model in older adults with MCI to identify AD risk factors on chromosome 19 that exacerbate cognitive decline |
| Presenter: | Jinhyeong Bae |
| Institution: | Indiana University – Purdue University, Indianapolis |
| Authors: | Jinhyeong Bae1, Kwangsik Nho1, Andrew J Saykin1, Angelina Polsinelli1, Dustin Hammers1, Kelly Nudelman1, Valentin T Pentchev2, Liana G Apostolova1 1 Indiana University School of Medicine, 2 Indiana University Network Sciences Institute |
Abstract
| Background/Significance/Rationale: | Research into genetic mapping possesses strong potential to inform the conversion from mild cognitive impairment (MCI) to Alzheimer’s disease (AD). We extended our previously developed novel deep learning framework to analyze MCI participants. The top 35 strongest AD-risk factors and their chromosomal risk impact score (CRIS), which indicates each SNP’s contribution in AD occurrence, determined by the model were utilized to characterize participants with MCI who were likely to convert to AD dementia (MCI-C) vs not (MCI-NC) over 3 years. |
| Methods: | The highest CRIS-ranked 35 AD-risk SNPs were utilized to differentiate MCI-C (n=203) and MCI-NC (n=213) participants enrolled in the Alzheimer’s Disease Neuroimaging Initiative. We predicted the rate of cognitive decline (memory, language, executive, and visuospatial function) in MCI using multiple regression with 5 SNPs’ CRIS as predictors. Lastly, we performed computational CRISPR to demonstrate the impact of SNP rs56131196 (APOC1), the strongest AD-risk SNP, in MCI-C participants. |
| Results/Findings: | SNPs in APOC1, TOMM40, and NECTIN2 showed significantly stronger CRIS for MCI-C than MCI-NC participants (p<0.001). All regression models predicting the rate of cognitive decline were significant (p<0.001). The r2-adjusted values were 0.279, 0.163, 0.098, and 0.178, for the memory, language, executive, and visuospatial models, respectively. MCI-C participants with the substitution of AA or AG genotype with GG were predicted to have a significantly lower likelihood of AD occurrence than those without substitution (p<0.001). |
| Conclusions/Discussion: | Our deep learning model trained on AD and CU participants successfully determined SNPs that predict conversion from MCI to AD dementia. |
| Translational/Human Health Impact: | Genetic screening based on regression models could be useful for patient selection in clinical trials with disease-modifying therapies. Furthermore, our computational CRISPR simulations in MCI-C confirm the significant promise of CRISPR for precision medicine. In vitro and in vivo animal and human studies exploring nucleotide-level substitutions are warranted to fully appreciate their role in translational neuroscience. |
