Wray- “proof of the pudding” - look at prediction in real world data. PRS accuracy varies considerably across samples (eg here in PGC depression) #WCPG23 🧬💻

Wray conceives of multi-PRS as borrowing information from other traits

Naomi Way discussant says that the new PRS methods all substantially outperform the simple weighted sum by flexibly taking into account different genetic architectures, but generally perform roughly similarly in psychiatric traits

Paul O’Reilly talks on using polygenic scores to infer genetic architecture in the tails of complex traits

… such as ADHD prediction improvement +++

Clara Albiñana presents multi-PGS using thousands of publicly available GWAS, resulting in massive performance improvement in some traits

Florian Privé describes LDpred2-auto , including three parameter model now, individual confidence intervals and inference, and future plans, but cautions more SNPs and more model parameters don’t help when underpowered

Session on polygenic risk score methods 🧬💻#WCPG23 Jian Zeng - using high density SNP coverage (7m) plus functional annotation SBayesRC shows consistent better performance on benchmarking