Machine Learning Prediction Model for Alzheimer's Disease (AD) Pathology Scores using multi-omics profiling and environmental factors

Alzheimer’s Disease (AD) is a progressive brain disorder with neuropathological changes and cognitive decline. Neuropathological changes, quantifiable by ß-amyloid (Aß) plaques and tau neurofibrillary tangles (NFT) are decisive for diagnosing the disease. Beyond well-known risk factors such as age, sex, and genetic background influencing AD, multi-omics profiling and environmental factors have been established to be marginally associated with Aß and NTF-related scores. These numerous factors make it challenging to exhaustively investigate the joint or relative predictive contribution of all factors in a single study. As some previous research studies investigated the marginal association of a limited set of factors with AD pathology scores, the others, because of the inaccessibility of human brain tissues, were totally based on animal samples. There is an urgent need to deeply understand the joint and relative contribution of these numerous factors in predicting AD for human patients. This project will address this question by conducting a cross-sectional study using brain tissue donors, environmental factors, and further AD-related factors to develop a machine learning predictive model for AD pathology scores. The model development will be based on a human brain database from the Emory University Goizueta AD Research Center (ADRC) with more than one thousand brain tissue donors. From these brain tissues, omics data, including DNA methylation and metabolomics data, have been assessed for about 260 participants. The brain bank has been enriched by many other factors, such as environmental factors, age, sex, anthropometric characteristics, genetic markers, and further markers related or potentially related to AD. While these collected markers are advantageous for understanding both joint and relative predictive contributions of AD-related factors for AD pathology scores, they impose the challenge of analyzing different types of data simultaneously. Random forests (RFs) have been shown to be a well-performing interpretable multi-omics predictive machine learning method. The proposed project will utilize the RF method to predict AD neuropathology scores using AD-related markers as predictor variables. The resulting model could clarify both the joint and relative predictive abilities of AD-related factors. Additionally, it may be used by clinicians to predict AD disease development.