Quantitative Issues in Cancer Research Working Seminar

Carmen Rodriguez Cabrera
Doctoral Student, Department of Biostatistics, Harvard T.H. Chan School of Public Health

Abstract: Personalized medicine has become very popular because it recognizes that individuals may respond differently to the same treatment due to differences in their underlying biology. With the emergence of large amounts of data coming from electronic health records linked with biobank specimens, large-scale health surveys, and environmental data, we have the opportunity to develop accurate and cost-effective personalized risk prediction models and methods that could ultimately have an impact on clinical practice. One example of such personalized risk prediction models is polygenic risk scores (PRS). Although easy to calculate, one big limitation of PRS is their transferability, that is their accuracy across many demographic groups that are underrepresented in precision medicine research. In fact, previous research has shown that most participants in existing large-scale genomic studies are of European ancestry. For instance, the UK biobank has more than 95% European ancestry (EA). Therefore, when trying to generalize PRS that are trained in mostly homogeneous genomic studies, their accuracy reduces with genetic distance from EA. Despite efforts to conduct GWAS with more diverse sub-populations, the gap remains in terms of sample sizes between EA and non-EA populations. Therefore, efforts have been made in statistical methodologies to try and address these disparities in PRS performance.  In this presentation, we aim to synthesize the statistical literature on single and multi-ancestry PRS estimation methods. Additionally, we will briefly introduce our preliminary idea/set up for a potential multi-ancestry PRS estimating method.