Enhancing Bayesian causal models using prior expertise from ontologies, for patients of Alzheimer's Disease. Lead Investigator: Hengyi Hu Institution : George Mason University E-Mail : hengyihu@gmail.com Proposal ID : 1582 Proposal Description: The research using this dataset has two primary aims: 1) to create a causal model of symptoms experienced by Alzheimer's patients using the Min-Max Hill Climbing algorithm, and 2) to improve this model using prior expertise found in authoritative medical ontologies, such as MedDRA, ICD-10-CM, and SNOMED CT. This is a novel method meant to go beyond what our current learning algorithms can provide us, and tap into a pre-existing source of expertise for a disease. The hypothesis is that the prior expertise in an ontology can be used to orient the causal Bayesian networks, creating a model that better fits the data. For example, in SNOMED CT, alcohol and the consumption of alcohol is a cause of anxiety. Knowing this causal relationship, the model learned from data can be constrained. Knowing several of these relationships would alter the structure of the network, and force the algorithm to consider prior expertise before fitting the data to the structure. We can compare these causal models, and confirm the changes of the structure using existing epidemiological literature. Successfully comparing the different models, we will have identified a way to create a better causal model that does not depend solely on the complexity of the algorithm. Instead, the model will be based on existing expertise and show that statistical methods and complex algorithms only take us so far when determining causality.