Novel Candidate Genes Detection Using Bayesian Network-Based Genome-Wide Association Study of Latent Traits in F2 Chicken Population

In chickens, economically important traits are commonly controlled by multiple genes and are often correlated. The genetic mechanisms underlying the correlated phenotypes likely involve pleiotropy or linkage disequilibrium, which is not handled properly in single-trait genome-wide association studies (GWAS). We employed factor analytical models to estimate the value of latent traits to reduce the dimensionality of the adjusted phenotypes. The dataset included phenotypes from 369 F2 chickens, categorised into six observable classes, namely body weight (BW), feed intake (FI), feed efficiency (FE), immunity (IMU), blood metabolites (BMB), and carcass (CC) traits. All birds were genotyped using a 60K SNP Beadchip. A Bayesian network (BN) algorithm was used to discern the recursive causal relationships among the inferred latent traits. Multi-Trait (MT) and Structural Equation Model (SEM) were applied for association analysis. Several candidate genes were detected across six phenotypic classes, namely the IPMK gene for BW and FI, and, the MTERF2 gene for BW and FE. The rs14565514 SNP, close to genes IPMK, UBE2D1, and CISD1, was recognised as a pleiotropic marker by both models. The NRG3 gene, located on chromosome 6, was associated with FI. CRISP2, RHAG, CYP2AC1, and CENPQ genes, located on chromosome 3, were detected for BMB through both MT- and SEM-GWAS. In general, the results indicated that the SEM-GWAS is superior to MT-GWAS due to considering the causal relationships among the traits, correcting the effects of the traits on each other, and also leading to the identification of pleiotropic SNP markers.