Abstract: Identifying genetic, environmental and demographic factors that underlie the risk of complex disease is extremely challenging. We present an explainable AI framework to untangle these relationships, one that facilitates identifying new predictors and quantifying their impacts on disease onset and progression. Our framework is built upon Bayesian Networks, which capture health factor interdependencies and provide completely explainable AI. We demonstrate the framework’s utility using diabetes as an exemplar disease.

Describe the new knowledge and additional skills the participant will gain after attending your presentation.: Participant will learn state-of-the-art approaches for EHR mining that can produce clinically actionable results for complex diseases (in particular diabetes).
Participant will be able to recognize the importance of Explainable Artificial Intelligence (XAI)(1–3) and how it can be used to discover novel interdependencies among health factors and determine confounding variables that are related to complex diseases and their comorbidities (in particular diabetes).
Participant will learn how to use our pre-trained Bayesian Networks to investigate his/her hypotheses on diabetes etiology and how to deploy it on his/her data that is not necessarily related to diabetes.

Authors:

Sergiusz Wesolowski (Presenter)
University of Utah

Gordon Lemmon, University of Utah
Alex Henrie, University of Utah
Edgar Hernandez, University of Utah
Jose Lazaro Guevara, University of Utah
Marcus Pezzolesi, University of Utah
Mark Yandell, University of Utah