Background: Until now, Mendelian randomization (MR) studies have investigated the causal association of risk factors with Alzheimer’s disease (AD) using large-scale AD genome-wide association studies (GWAS), GWAS by proxy (GWAX), and meta-analyses of GWAS and GWAX (GWAS+GWAX) datasets. However, it currently remains unclear about the consistency of MR estimates across these GWAS, GWAX, and GWAS+GWAX datasets. Methods: Here, we first selected 162 independent educational attainment genetic variants as the potential instrumental variables (N = 405,072). We then selected one AD GWAS dataset (N = 63,926), two AD GWAX datasets (N = 314,278 and 408,942), and three GWAS+GWAX datasets (N = 388,324, 455,258, and 472,868). Finally, we conducted a MR analysis to evaluate the impact of educational attainment on AD risk across these datasets. Meanwhile, we tested the genetic heterogeneity of educational attainment genetic variants across these datasets. Results: In AD GWAS dataset, MR analysis showed that each SD increase in years of schooling (about 3.6 years) was significantly associated with 29% reduced AD risk (OR=0.71, 95% CI: 0.60–0.84, and P=1.02E−04). In AD GWAX dataset, MR analysis highlighted that each SD increase in years of schooling significantly increased 84% AD risk (OR=1.84, 95% CI: 1.59–2.13, and P=4.66E−16). Meanwhile, MR analysis suggested the ambiguous findings in AD GWAS+GWAX datasets. Heterogeneity test indicated evidence of genetic heterogeneity in AD GWAS and GWAX datasets. Conclusions: We highlighted significant difference and genetic heterogeneity in clinically diagnosed AD GWAS and self-report proxy phenotype GWAX. Our MR findings are consistent with recent findings in AD genetic variants. Hence, the GWAX and GWAS+GWAX findings and MR findings from GWAX and GWAS+GWAX should be carefully interpreted and warrant further investigation using the AD GWAS dataset.

The rs6647 variant G allele in SERPINA1 gene was reported to be associated with the risk of large artery atherosclerotic stroke (LAS), however, the mechanism remains unclear. Here, we performed a functional annotation of the rs6647 variant by using the software HaploReg version 4.1 (HaploReg v4.1). Next, the expression quantitative trait loci (eQTLs) analysis of multiple datasets was conducted for determining the association between the rs6647 and SERPINA1 expression in various tissues. Then, a case‐control gene expression analysis was done using two independent ischemic stroke (IS) gene expression datasets. Finally, SERPINA1 expression in whole blood samples from 8 LAS patients and 14 healthy persons were compared. The functional annotation suggested that the rs6647 regulates gene expression in multiple human tissues especially in brain and blood. The eQTLs analysis revealed a significant association of the rs6647 G allele with increased expression of SERPINA1 gene only in whole blood. Compared with the controls, there was an increased expression of SERPINA1 gene in whole blood in both IS patients and LAS patients. SERPINA1 gene expression in whole blood bridges the rs6647 variant G allele with increased LAS risk, providing new insights into the mechanisms underlying role of the rs6647 in determining LAS risk.

Multiple sclerosis (MS) is an autoimmune disease for which it is difficult to find exact disease-related genes. Effectively identifying disease-related genes would contribute to improving the treatment and diagnosis of multiple sclerosis. Current methods for identifying disease-related genes mainly focus on the hypothesis of guilt-by-association and pay little attention to the global topological information of the whole protein-protein-interaction (PPI) network. Besides, network representation learning (NRL) has attracted a huge amount of attention in the area of network analysis because of its promising performance in node representation and many downstream tasks. In this paper, we try to introduce NRL into the task of disease-related gene prediction and propose a novel framework for identifying the disease-related genes multiple sclerosis. The proposed framework contains three main steps: capturing the topological structure of the PPI network using NRL-based methods, encoding learned features into low-dimensional space using a stacked autoencoder, and training a support vector machine (SVM) classifier to predict disease-related genes. Compared with three state-of-the-art algorithms, our proposed framework shows superior performance on the task of predicting disease-related genes of multiple sclerosis.