Cross-Disorder Machine Learning Uncovers Schizophrenia Risk Variants Predictive of Alzheimer's Disease

Abstract

Alzheimers disease (AD) and Schizophrenia (SCZ) exhibit overlapping clinical features and biological mechanisms, but the extent of their shared genetic etiology and the potential for cross-disorder risk prediction are not fully elucidated, with previous genetic studies yielding mixed results. This study investigated whether integrating statistically-selected SCZ-associated single nucleotide polymorphisms (SNPs), identified from large-scale GWAS, could enhance machine learning (ML)-based prediction of AD risk beyond models using only AD-associated SNPs. Utilizing summary statistics from large-scale GWAS for AD (N=1,126,563) and SCZ (N=320,404), alongside reference genotypes from the Alzheimers Disease Sequencing Project (ADSP, N=7,416 Non-Hispanic Whites), we employed an objective iterative ML framework evaluating distinct classification algorithms and adding prioritized SCZ SNPs to a baseline AD-SNP model. Integrating the top 50 SCZ SNPs significantly improved AD prediction accuracy in the best-performing ML model, increasing the area under the receiver operating characteristic curve (AUC) from 0.6032 (baseline) to 0.6400 (DeLongs test p=0.0005). Contributing SCZ SNPs implicated shared biological pathways relevant to AD, including immune regulation/neuroinflammation, tau protein biology, and synaptic vesicle trafficking, while also revealing novel predictive variants warranting further investigation. These findings demonstrate that incorporating specific SCZ genetic risk variants can modestly but significantly enhance AD risk prediction, supporting meaningful genetic overlap and providing novel genetic targets for cross-disorder research, highlighting the potential of this approach for dissecting the complex genetic architecture of brain disorders.