Integrative machine-learning on high-throughput human data identifies age-specific hallmarks of Alzheimer’s disease

Alzheimer’s disease (AD) is an incredibly complex and presently incurable age-related brain disorder. To better understand this debilitating disease, we collated publicly available RNA-Seq, microarray, proteomics, and microRNA samples derived from AD patients and non-AD controls. 4089 samples originating from brain tissues and blood remained after applying quality filters. Since disease progression in AD correlates with age, we stratified this large dataset into three different age groups: < 75 years, 75-84 years, and ≥ 85 years. The RNA-Seq, microarray, and proteomics datasets were then combined into different integrated datasets. Ensemble machine learning was employed to identify genes and proteins that can accurately classify samples as either AD or control. These predictive inputs were then subjected to network-based enrichment analyses. The ability of genes/proteins associated with different pathways in the Molecular Signatures Database to diagnose AD was also tested. We separately identified microRNAs that can be used to make an AD diagnosis and subjected the predicted gene targets of the most predictive microRNAs to an enrichment analysis. The following key themes emerged from our machine learning and bioinformatics analyses: cell death, cellular senescence, energy metabolism, genomic integrity, glia, immune system, metal ion homeostasis, oxidative stress, proteostasis, and synaptic function. Many of the results also demonstrated unique age-specificity. For example, results highlighting cellular senescence only emerged in the earliest and intermediate age ranges while the majority of results relevant to cell death appeared in the youngest patients. These data demonstrate that, like aging, AD is a multifaceted process characterized by diverse dysfunction. Please see paper for detailed methods.

阿尔茨海默病（Alzheimer’s disease, AD）是一种极为复杂且目前尚无治愈方案的年龄相关性脑部疾病。为深入解析这种致残性疾病，我们汇聚了公开可得的、来自AD患者与非AD对照个体的RNA测序（RNA-Seq）、微阵列（microarray）、蛋白质组学（proteomics）以及微小RNA（microRNA）样本。经过质量过滤后，最终保留了4089份来源于脑组织与血液的样本。鉴于AD的疾病进展与年龄密切相关，我们将该大型数据集划分为三个年龄组：75岁以下、75至84岁以及85岁及以上。随后将RNA测序、微阵列及蛋白质组学数据集整合为多组联合数据集。我们采用集成机器学习（ensemble machine learning）方法，筛选出可准确将样本分类为AD患者或对照个体的基因与蛋白质。随后对这些预测性特征开展基于网络的富集分析。我们还测试了分子特征数据库（Molecular Signatures Database）中不同通路相关基因/蛋白质的AD诊断效能。我们另外筛选出可用于AD诊断的微小RNA，并对预测性最优的微小RNA的靶基因开展富集分析。本研究的机器学习与生物信息学分析揭示了以下核心主题：细胞死亡、细胞衰老、能量代谢、基因组完整性、神经胶质细胞（glia）、免疫系统、金属离子稳态、氧化应激、蛋白质稳态以及突触功能。多数研究结果还呈现出显著的年龄特异性。例如，聚焦细胞衰老的分析结果仅在早期和中期年龄组中出现，而与细胞死亡相关的多数结果则见于最年轻的患者队列。本研究数据表明，与衰老类似，AD是一种以多种功能异常为特征的多维度病理过程。详细实验方法请参见原文。