Association analysis between polygenic risk scores and traits: practical guidelines and tutorial with an illustrative data set of schizophrenia

Abstract

Most methodological Polygenic Risk Score (PRS)-related papers explain the laborious process of computing the PRS in great depth. Afterwards, as a last step, it is generally described that to test a possible association between a PRS and a trait of interest, an analysis through regression models (linear or logistic, depending on data type) should be carried out adjusting for covariates (e.g., sex, age, clinical information, or genetic ancestry-based Principal Components). When covariates are included, measurements such as the increment on the variance explained by the addition of the PRS to the model or the significance of the PRS term are usually reported. However, the association study between PRSs and a trait is a complex concern that requires proper modeling and analysis, since interactions and validation conditions represent crucial aspects. Even though excellent papers explain how to use and interpret the results obtained with such regression models, sometimes important information from the previously calculated PRS may be lost, partly due to the automation of analyses. With this guide, we intend to fill a gap in association studies between PRSs and a trait and to facilitate the analysis, obtaining statistically correct results. It contains a motivating real data case analyzed exhaustively to illustrate how to face a real analysis. Besides, it is accompanied by four examples, called Working Examples, which present different situations the researcher may encounter along with the R code for analyzing all these data sets and the corresponding application of the steps in this guide.