Gene Profile Analysis of Associations between Alzheimer's Disease and Circadian Rhythmic Patterns of Human Brain Regions Using Machine Learning

Alzheimer’s disease (AD) is the leading type of dementia and has an increasing death rate. It is observed that AD patients have disrupted circadian patterns, and recent studies have also shown emerging evidence that restoring the circadian rhythm of a brain region is a potentially effective treatment strategy. Although the close relationship of AD with circadian rhythms has been studied for several individual genes, there are no functional associations of AD with rhythmic patterns of various regions of the human brain at the gene-profile level to tailor potential treatment or research design. This problem is exacerbated by the inability, until now, to obtain gene expression data with appropriate timestamps of different regions in the human brain from a cohort of AD patients of reasonable size. The increased amount of gene expression data of various brain regions of AD patients, without timestamps, has not only empowered the identification of genes as risk factors for AD, but created an opportunity to use these untimed data to analyze gene profiles to discover functional links between AD and the tissue-level circadian rhythms. Machine learning (ML), a form of artificial intelligence, will be used to examine the gene profiles of tens of thousands of genes simultaneously for various regions of the human brain. Rather than analyzing each clock gene individually, which may have the same circadian cycles yet different phases, we are using ML to explore a multifactor analytic model in an intrinsic, latent, low-dimensional subspace of all the genes to generate the most accurate algorithm to identify the rhythmic patterns and associations of different brain regions with AD. Our overall aim is to discover circadian rhythmic patterns based on gene profile analysis, and leverage them to identify and understand the associations between AD and circadian rhythms of different brain regions. This project is a retrospective study based on the secondary analysis of the existing untimed gene expression database of randomized cohorts of AD patients. Specific Aim 1: Discover circadian rhythmic patterns to further identify the associations between AD and circadian rhythms of different brain regions using machine learning. In this Aim, we will develop algorithms to achieve the following capability: 1. Gene expression data sets of various brain regions of a cohort of subjects with or without AD will be merged and analyzed by ML to create an algorithm for deciphering rhythmic patterns of different regions. This method will be designed to explore the overall rhythmic effects of the genes at the tissue level. 2. AD is closely related to circadian rhythms of the human brain, and in this Aim, we will examine these patterns to determine associations between circadian rhythms of brain regions for AD patients and normal controls in our data set. ML will be used to create an inter-region link discovery algorithm for circadian rhythms. These algorithms will be developed by accounting for potential gender and ethnicity factors. Vital to the success of any discovered associations is usability and reproducibility. In this Aim, we will partially verify our algorithms over mouse brain gene expression data with timestamps, which will be used for transfer learning as well as calibrate the algorithms. We will perform audited oversight to ensure reproducible translation of the methodology for Aim 2. Milestone will be to create an intermediate validation that can attain high-quality association results with reproducibility through the exploratory human brain data and the oversighting mouse brain data. Specific Aim 2: Use a database of different cohorts to verify the applicability and validity of discovered patterns and associations of circadian rhythms with AD. We will apply the developed method to discover patterns from genomic data of different cohorts. In a blinded fashion, we will use the developed algorithm to discover the rhythmic patterns out of gene profiles, and then identify their associations with AD. This Aim will confirm whether our algorithm is able to identify rhythmic patterns and associations with AD not only in the AMPAD population but in other diverse populations. The platform for finding the patterns and associations with AD will be optimized for our gene profile analysis. Methods validation will be carried out to determine the reliability sample size and analyses across cohorts as well as add additional cohorts. Anticipated outcomes: For the first time, gene profile analysis based on untimed expression data of tens of thousands of genes will be performed to pinpoint the AD-related rhythmic patterns of different brain regions. To our knowledge, this is the only capacity of its kind, which permits the most direct possible glimpse into the rhythmic events in a human brain. This method will be used to find rhythmic patterns and correlations between brain regions, and further compare them for subjects with or without AD. Identification of the patterns and correlations should advance our understanding of the relationship between brain regions’ circadian rhythms and AD, and also provide researchers with a potentially critical tool for developing treatment of AD.