An Interpretable and Inclusive AI Framework for COVID-19 Biomarker Discovery and Risk Prediction

Coronavirus Disease 2019 (COVID-19) continues to be a public health concern, with some patients experiencing mild or no symptoms, while others require hospitalization or mechanical ventilation. Research into the genetic differences among COVID-19 patients has uncovered genes linked to disease severity. However, these genes explain only a small portion of the variation in COVID-19 severity and have limited predictive power. Translating these findings into practical knowledge is difficult because current methods fall short of identifying causal genes and often overlook interactions between them. While artificial intelligence (AI) methods can model complex relationships between genes and show high predictive accuracy, their clinical use is limited due to their "black box" nature. In this study, we will develop an inclusive and interpretable AI tool to predict COVID-19 severity using genetic and demographic data. Using data from a Canadian COVID-19 cohort, we aim to develop an AI model that identifies genes linked to COVID-19 severity. We will use statistical methods to reduce errors and better understand how the model makes decisions. Next, we will incorporate genetic and demographic factors to construct an AI model to estimate an individual's risk of developing severe COVID-19. To ensure equitable performance across ancestries, sex, and gender, we will utilize innovative AI techniques that allow our models to learn from diverse populations and apply those insights effectively to others. This approach will help us address potential disparities and enhance inclusivity in our analyses. Furthermore, we will benchmark our model against existing state-of-the-art approaches and validate its performance using data from British and American COVID-19 cohorts. The results of our work will shed light on the genetic and biological mechanisms underlying COVID-19 severity and may improve vaccine development, individual patient outcomes, and healthcare decision-making.