Alzheimer’s disease (AD) is defined by the presence of amyloid plaques and tau tangles, but in reality, distinguishing AD from other neurodegenerative diseases is surprisingly difficult. Over half of AD patients have brain changes typical of other neurodegenerative diseases, which creates an incredibly complex picture when trying to diagnose and treat patients. Understanding how these different pathologies interact and influence disease progression is challenging, given the many factors at play and how they affect each other. Dr. Stein-O’Brien is an expert in developing computational tools to understand these interactions and aims to deploy them to identify the disease mechanisms underlying each specific case and combination of pathologies.
This project focuses on implementing three computational tools to investigate neurodegenerative disease mechanisms at the Johns Hopkins Brain Resource Center, which includes over 3,000 brain samples. The first tool to be used is Multi-Omic Regional Probabilities Honed via non-negative matrix factorization (MorphNMF). MorphNMF is an algorithm designed to interpret different forms of spatial-omic techniques, including proteomics and transcriptomics. It incorporates both discrete variables, such as whether a case is tau-positive or negative, and continuous variables, such as neuron size, shape, and distribution. The first aim of the project will be to deploy MorphNMF and incorporate other information sources, including neuronal activity measurements, imaging data, and clinical information. The goal is to encode and discover the regulatory relationships across omics and modalities that underlie different neurodegenerative processes.
The second tool is called InterPLatent. InterPLatent is a transformer; the same neural network architecture used in large language models like ChatGPT. InterPLatent is designed to help researchers study latent factors, which are variables that can’t be measured directly but can be deduced from the data, like inferring someone’s stress level from their heart rate and sleep patterns. In this case, it will be used to interpret interactions among latent factors inferred from different sources of pathological, genetic, protein, and clinical data. Unlike ChatGPT, which was trained on a huge corpus of data, the amount of clinical and molecular data remains quite small. Thus, to make this tool usable, the Stein-O’Brien lab will design it to operate with the dataset sizes and computational power available to most biomedical research labs.
The final tool to be developed in this proposal is called PhysiBrain. PhysiBrain will be a powerfully detailed simulation of disease progression based on the data sources collected and interpreted via the first two aims. By incorporating all data about cellular phenotypes and their interactions, the platform will enable testing of hypotheses in silico (through computer simulations) related to the cell interactions that contribute to different neurodegenerative disease processes. The lab believes this tool will allow them to study how various changes in brain cells unfold over time and what effects they trigger. This kind of time-based tracking is not possible in post-mortem human brain tissue.
This project seeks to develop and refine three powerful computational tools to understand and interpret the complex interplay among pathologies across different neurodegenerative diseases. By leveraging advances in computational power and techniques, Dr. Stein-O’Brien aims to transform our understanding of neurodegenerative diseases and how we model and study them.
