Artificial intelligence to create equitable multi-ethnic polygenic risk scores that improve clinical care

Overall Objective: To develop state-of-the-art artificial intelligence (AI) methods which address the ethnic inequities inherent in genomic medicine and rapidly emerging polygenic risk scores (PRSs). We will subsequently apply these approaches to demonstrate improved screening, diagnosis and treatment of common disease. Background: Recent breakthroughs in genomics and machine learning have generated personalized medicine tools that have the potential to improve patient care and health maintenance through PRSs. PRSs are risk predictors, which combine information across the entire genome to predict a person's risk of disease, or whether they have unfavorable disease risk factors, such as high cholesterol levels. While still evolving rapidly, these PRSs are more predictive of susceptibility to disease than many traditional disease risk factors. Since PRSs can predict risk of disease they could be helpful in screening programs, by identifying a group of individuals at such low risk of disease, that screening would be unlikely to be helpful-thereby allowing screening programs to focus on individuals at highest risk. They could also refine diagnosis, by identifying people in the population most likely to have a disease and then targeting required diagnostic testing in those at highest risk. Last, they can also improve disease treatment by identifying individuals most likely to benefit from treatment. While PRSs could help to improve clinical care, they have been largely developed in individuals of European-only ancestry. This is because the large cohorts from which PRSs have been developed consist predominantly of European-ancestry participants. Since genetic make-up varies by ancestry, the performance of these PRSs in non-European ancestries is considerably worse; it is known that not only differences in in genetic risk factors lead to attenuation of performance of PRSs, but also that changes in correlation patterns in our genomes due to distinct population histories decrease predictive accuracy even if the actual genetic risk factors remain the same. This creates several important problems for the roll-out of such tests in the Canadian and UK healthcare environments. This is because approximately 22% of Canada's population and 12% of the UK are visible minorities. Further, these populations can have increased rates of health care utilization. Therefore, use of European-only PRSs could serve to worsen existing health disparities. Canada's Chief Information Officer stated that, "Using Artificial Intelligence in government means balancing innovation with the ethical and responsible use of emerging technologies." Using PRSs developed to benefit only the majority group in our societies would not respect these directives and therefore our program will serve to ensure the responsible use of AI. What are our Specific Goals? 1) To develop AI methods and open source software packages to improve the accuracy of PRSs for CHD and hyperlipidemia for individuals of non-European ancestry. 2) To compare the performance of these ancestry-adapted PRSs in individuals in a diverse set of cohorts, representing multiple ancestries in the UK and Canada. The Team: Recruiting leaders across the UK and Canada, we have a depth of expertise in AI methods development, human genetics, clinical medicine, cohort development for minority groups, and statistical genetics. This gender-balanced team involves an appropriate mix of senior and junior investigators. Relevance: Success in this AI-enabled program will allow for the transfer of recent advances in genomic medicine to the citizens of Canada and the UK, regardless of ancestry. The AI methods developed will be widely applicable to other PRSs in development by other groups. As genomics takes root in regular clinician-patient interactions, our set of AI-based tools will ensure that the benefit derived from these advances will be shared by all citizens of our countries.