Approved Research

Deep learning with model interpretation and uncertainty quantification for polygenic risk scores and genome wide association studies

Lay summary

Complex traits and diseases are often impacted by one's genetic composition. And predictive genomics provides a method to gain insight into how genetic variations can be used to estimate genetic risk for complex traits and diseases. Our project aims to enhance the accuracy and reliability of genetic predictions across a range of traits and diseases. We will develop, train, and test machine learning models designed to predict risk for these complex traits and diseases. In addition, we will perform model interpretation to understand why a model predicts the way it does. This will allow for the identification of significant features used in the model's decision-making process. We also seek to dive deeper to understand and communicate how confident we can be in the predictions through uncertainty quantification techniques. By systematically quantifying uncertainty, we aim to enhance the trustworthiness of genetic risk assessments, critical for informed medical decision-making.

We anticipate this project will last approximately three years from accessing the data to publication of results. The public interest is served through more accurate and reliable genomic predictions, empowering healthcare providers with tools to provide personalized care plans, ultimately improving patient outcomes and advancing the broader understanding of genomics in the public health domain.