Deep learning to enable the genetic analysis of aorta

Project Summary / Abstract Aortic disease is an important contributor to cardiovascular morbidity and sudden death. Key discoveries, including identification of the causal gene for Marfan’s syndrome (FBN1), have advanced our knowledge of syndromic aneurysm and dissection, but to date there remains insufficient information on sporadic thoracic aortic disease. For example, despite growing knowledge of the importance of aortic disease, there is no guideline for screening for ascending aortic disease, and no therapy to treat its underlying molecular mechanisms. While there is likely some overlap between thoracic and abdominal aortic disease, they are embryologically distinct and likely have different genetic and clinical risk factors. In Dr. Pirruccello’s preliminary work, he developed an automated deep learning model to quantify the diameter of the thoracic aorta using cardiovascular magnetic resonance imaging (MRI). He applied the model in the UK Biobank and conducted a genome-wide association study for the diameter of ascending and descending thoracic aorta in nearly 40,000 participants. These results cemented the feasibility of the approach of (1) training deep learning models to extract biologically relevant information from imaging, and (2) conducting genetic analyses on these deep learning model-based phenotypes. This now paves the way for a more comprehensive analysis of additional aortic traits, and downstream evaluation of genetic risk factors for both thoracic and abdominal aortic disease. First, Dr. Pirruccello proposes to develop models for additional aortic traits including thoracic aortic strain and distensibility, and abdominal aortic diameter. Second, after developing additional models to extract those features, Dr. Pirruccello proposes to conduct genetic analyses on these traits in the UK Biobank, elucidating the common and rare genetic variation that leads to variability in the aorta’s size and distensibility at several levels. Third, he proposes to produce polygenic scores, permitting modeling of the clinical and genetic risk for abnormalities in aortic size and distensibility that may predispose to aortic aneurysm and dissection. This work will take place in the Division of Cardiology at the Massachusetts General Hospital, and at the Broad Institute of MIT and Harvard. Dr. Pirruccello will perform this research under the mentorship of Dr. Patrick Ellinor, the Director of the Cardiovascular Disease Initiative at the Broad Institute, and Dr. Mark Lindsay, an expert in genetic aortic disease at the Massachusetts General Hospital Thoracic Aortic Center. Dr. Pirruccello’s goal is to become a computational cardiovascular geneticist with expertise in machine learning. He is dedicated to becoming an independent investigator and to use the research performed for the K08 as a springboard for an R01.