PRINCIPAL COMPONENTS ANALYSIS CORRECTS COLLIDER BIAS IN POLYGENIC RISK SCORE EFFECT SIZE ESTIMATION

BACKGROUND: Genome-wide polygenic scoring has emerged as a way to predict psychiatric and behavioral outcomes and identify environments that promote the expression of genetic risks. An increasing number of studies demonstrate that the effects of polygenic risk scores (PRS) may be biased by the inclusion of heritable environments as covariates when the environment is influenced by unmeasured confounding variables, an example of collider bias. Inclusion of the principal components of observed confounders as covariates may correct for the effect of unmeasured confounders. METHODS: A simulation study was conducted to test principal components analysis (PCA) as a correction for collider bias. Data were sampled from a model which tested different values for the effect of the polygenic risk score on the heritable environment, the correlation structure of the unmeasured confounding data, and the proportion of the confounding data that is used to construct the principal components. Other model parameters were fixed across all simulation iterations. RESULTS: Modeling the first PC of observed confounders as a covariate recovers the PRS effect size estimate under reasonable assumptions about the proportion of the confounding data that is measured or the correlation structure of the confounding data. Required assumptions become stricter as the effect of PRS on environment (and the magnitude of bias) increases. CONCLUSION: Inclusion of the first PC of observed confounders as a covariate may improve the accuracy of PRS effect size estimation when heritable environments are included in the model as covariates. Future directions include application of this method in observed data.https://scholarscompass.vcu.edu/gradposters/1130/thumbnail.jp