Computational Modelling and Inference of Neurodegenerative Disease Propagation

This project uses computational modelling, machine learning, and big-data analysis to address key unknowns on the biology of neurodegenerative disease: when and where it starts; how it spreads over the brain (“propagates”); how it varies among diseases, subtypes, and individuals; how risk factors influence mechanisms. Current models linking imaging and other clinical data to propagation mechanisms are crude, and evaluation remains largely qualitative. The project introduces a powerful Bayesian inference framework, accounting for uncertainties throughout the data-processing pipeline, to enable robust evidence quantification of new and detailed computational models of disease propagation. The approach demands a rethink of contributing technologies. First, we unravel patient heterogeneity, which confounds propagation-model evaluation, through a new generation of data-driven disease progression models that identify fine-grained disease subtypes characterised by temporal trajectory of multi-modal biomarkers. Second, propagation models use brain connectome estimates and are confounded by high connection-error rates. We establish a new probabilistic connectomics paradigm that quantifies likelihoods of false positive and negative connections enabling models to mitigate their inevitable presence. We use the framework to provide the first truly quantitative evaluation of propagation models against state-of-the-art data sets leading to fundamental new insights on the mechanisms of pathology propagation in neurodegenerative diseases.