Research Question
We seek to obtain causally-faithful multi-modal counterfactuals of Alzheimer’s Disease (AD) patients.
A variety of factors can contribute to or are correlated to the development of Alzheimer’s Disease: genetic such as the APOE-epsilon4 gene, lifestyle oriented such as obesity and diabetes, or brain volume shrinkage, information extractable from the UK Biobank. Each data modality uncovers partial information of these factors, and they interact in a causal manner to each other, leading to a chain of causations. As an example, time plays a confounding role in AD, parts of the brain (e.g. Hippocampus) are known to shrink with time, whilst also the accumulation of biomarkers such as Amyloid-beta plaques in the brain seem to eventually lead to AD.
The complex relationships between these factors can be captured through causal inference. It can then be embedded into AI tools to create a causal AI model of AD, capable of generating causally-faithful counterfactuals, data never obtainable in the real world, facilitating discovery of new causal biomarkers.
Objectives
We first aim to build a causal graph that captures variables that are the most important in the development of AD, including time (the patient’s age). We then embed this graph into AI tools to obtain counterfactuals, which can then be exploited for e.g. biomarker discovery.
Scientific Rationale
The world population is becoming older, while the age-standardised prevalence of AD remains largely the same, leading to an explosion in the worldwide number of AD patients. In the United States alone, six million people were diagnosed with AD or mild cognitive impairment by 2017, and this number is forecasted to rise to 15 million by 2060. Discovery of new biomarkers via counterfactuals and understanding the causal processes leading to AD will allow acceleration in treatments to cure or even delay AD. This can then dampen the explosion of AD incidence worldwide.