Fig. 1: Schematic diagram of the study design.
From: Improving multi-trait genomic prediction by incorporating local genetic correlations
a Models for genomic prediction. STGBLUP: single trait GBLUP; MTGBLUP: multi-trait GBLUP; LGC-model-1 ~ 3: three models incorporating local genetic correlations (LGC). LGC-model-1 partitions the term \([\begin{array}{c}{{{{\bf{a}}}}}_{1}\\ {{{{\bf{a}}}}}_{2}\end{array}]\) (vector of genetic values (GV) for trait 1 and trait 2) in MTGBLUP into two parts, i.e., GV contributed by regions with significant LGCs and regions without significant LGCs. LGC-model-2 partitions \([\begin{array}{c}{{{{\bf{a}}}}}_{1}\\ {{{{\bf{a}}}}}_{2}\end{array}]\) into three parts, i.e., GV contributed by regions with strong positive LGCs, regions with strong negative LGCs, and the rest regions. LGC-model-3 accounts for LGCs by simply adjusting the SNP effects for a trait estimated from a single trait model (STGBLUP) using LGC-weighted SNP effects for the other trait. b Simulated data and three real datasets from human, cow, and pig populations for evaluating the performances of these models. c Estimating global genetic correlations between traits in the real datasets using GREML. d Estimating local genetic correlations using LAVA and GWAS summary statistics. e 10-fold cross-validation for evaluating the accuracies of genomic prediction.
