SNPs. PRS were generated at multiple P-value thresholds (PT) (at intervals of 0.01 ranging from P = 0.0001 to P = 1). Best-fit PRS (at PT = 0.3 for PTSD and PT = 0.3 for re-experiencing symptoms, respectively) were used to predict PTSD status under logistic regression, adjusting for 5 PCs and dummy study indicator variables, using the glm function in R 3.2.1. PRS prediction plots were based on quintiles of PRS, with odds ratios calculated in reference to the lowest quintile. The proportion of variance explained by PRS was estimated as the difference in Nagelkerke’s R2 between a model including PRS plus covariates and a model with only covariates. R2 was converted to the liability scale assuming a 30% prevalence, using the equation found in Lee et al89. P-values for PRS were derived from a likelihood ratio test comparing the two models.
