Combining machine learning and statistics to understand the genetic architecture of MRI-derived anatomical structures and develop disease risk models using these associations.
Approved Research ID: 87255
Approval date: November 7th 2022
We aim to combine machine learning and genomics to understand how changes in our genome impact the anatomy of the cardiovascular system (heart and blood vessels), and how this can lead to cardiovascular diseases (CVD). The diagnosis of CVD relies on the identification of clinical symptoms that can take years to manifest, by which time the disease is often severe and fatal. These symptoms often reflect anatomical damage accrued over time in organs and tissues throughout the body. Some of this anatomical damage can be diagnosed by body imaging such as Magnetic Resonance Imaging (MRI). While MRI diagnosis is better than waiting years for CVD symptoms to appear, it is still not soon enough to capture the disease at the very early stage. We hypothesize that gene variants that contribute to the anatomy of the cardiovascular system can be leveraged to predict risk for CVD. Here, we will use machine learning to extract anatomical features of the heart and blood vessel from cardiac, abdominal, and brain MRIs from healthy individuals and individuals with CVD. We will then correlate these features to the rich genomic data of the UK biobank. This comparison will reveal the extent to which genetic variation affecting the anatomy of the cardiovascular system can help better understand and predict the risk of CVD such as aneurysms, heart failure, and stroke.
Genomic, machine learning, risk prediction, statistical modeling
Our second project aims to identify genes that predispose to autism spectrum disorder (ASD) and congenital heart defects (CHD). CHD and ASD are the two most common diseases among children. Their high heritability suggests a strong genetic contribution, yet the underlying genes remain unclear. To tackle this question, we concentrate on regions of the genome where variants that affect the 3D structure of the genome have biological consequences such as changes in the regulation of nearby genes. We have developed genomic tools that can predict the 3D configuration of the genome and can identify regions in the genome that are jointly transmitted from parent to children more often than would be expected by chance alone. We will quantify how these regions contribute to the risk of ASD and CHD. Ultimately, this work will reveal genome variants that could be predictors of disease.
We anticipate that our project will take 3 years to complete. Our findings will significantly advance our understanding of genetic risk for diseases and help develop models for risk prediction and therapies beneficial for patients.