With the aim at improving the PRS prediction we attempted to exploit the genetic overlap of depression with other phenotypes. We therefore combined all the above PRSs into a single score as a weighted sum15. We chose here to use the log(OR) for the logistic regression of sub-phenotype of interest on each score as a continuous factor as a measure of ‘importance’ in context of the sub-phenotype of interest (either recurrent vs. single-episode or depression with additional diagnosis of bipolar disorder, schizophrenia or SUD). The logistic regressions were adjusted for the first 10 PCs of the second PCA mentioned above. We added each standardized PRS weighted by its log(OR) one PRS at the time to the aggregated score. We started with the phenotype with highest abs(log(OR)) and ended with the lowest. This way we ended up with a sequence of scores starting with S0 and continuing with: Sk=∑i=1klog(ORPi)SPi−μ(∑i=1klog(ORPi)SPi)σ(∑i=1klog(ORPi)SPi)
