Probabilistic PRS-based stratification framework.

Large uncertainty in individual PRS estimation impacts PRS-based risk stratification

Complex diseases occur as a result of many genes interacting with environmental influences (such as diet, sleep, stress and smoking). They are also called “polygenic” diseases - with “poly” meaning many and “genic” involving genes. Genes and variants associated with such complex diseases are identified by comparing the genomes of individuals with and without those diseases. The associated variants are then used in a “polygenic risk score (PRS)” to estimate how the collection of a person’s variants affect their risk for a certain disease.

Several studies have shown PRS are useful in predicting disease risk, forecasting disease trajectories and stratifying patients into different treatment groups for multiple diseases such as breast cancer, coronary artery disease and type 2 diabetes. Given its potential utility in personalized medicine, it’s essential to assess the bias and variance of PRS estimates for responsible use in individual level clinical decision making. However, the uncertainty in any given individual’s PRS estimate has not been explored yet.

In their recent work in Nature Genetics, Ding, Hou et al. proposed a general framework to estimate the uncertainty of an individual's PRS estimate using Bayesian approach. Unlike existing methods that provide a single estimate, this approach yields a credible range for the PRS which covers the true genetic value with 95% probability. In analyses of 13 complex traits from UK Biobank, they observed large credible ranges for any individual genetic value which impacts the interpretation of PRS-based risk stratification. For example, only 0.8% of the individuals with their PRS estimates in the top decile have their 95% credible interval fully contained in the top decile. To account for such high uncertainty they propose a probabilistic stratification framework that can aid personalized clinical decision making under various cost-benefit scenarios. Overall, this new work demonstrated the individual level PRS uncertainty in multiple phenotypes and diseases, emphasized the importance of taking it into account in individual level risk interpretation and stratification and proposed a probabilistic PRS-based stratification to account for uncertainty in decision making. 

Yi Ding, Kangcheng Hou, Kathryn S. Burch, Sandra Lapinska, Florian Privé, Bjarni Vilhjálmsson, Sriram Sankararaman, Bogdan Pasaniuc

Media Contact: 

Leticia Ortiz | Marketing & Communications | Building a community around data science in biomedicine​