#15 · Genomics

Transfer, kernel and penalized methods raise power in genetic association studies

The common thread is using auxiliary information or flexible penalized and kernel models to raise power in genetic and epigenomic association and prediction. tranScore transfers EUR-EAS summary statistics via genetic similarity to strengthen East Asian GWAS. GWKBR adds GWAS-informed priors to a Bayesian kernel model for genomic prediction, and GEiPRS models genotype-environment interactions in penalized polygenic risk scores. Spectral components from identity-by-descent graphs adjust for fine-scale structure, outperforming PCs in UK Biobank. Kernel multi-marker accelerated failure time tests handle left truncation and competing risks, while waveome applies Gaussian-process variable selection to longitudinal omics counts. The new element is adaptation to ancestry, survival, and temporal structure rather than standard single-marker designs.