Computational protocol: Performance Gains in Genome-Wide Association Studies for Longitudinal Traits via Modeling Time-varied effects

Similar protocols

Protocol publication

[…] Under the framework of the random regression model, we proposed two detection methods for the association analysis of longitudinal traits, i.e., functional GWAS model treating each SNP as the covariate (fGWAS-C), and functional GWAS model treating each SNP as the factor (fGWAS-F). To exploit the property of these two novel methods, several alternative models/strategies which used the EBVs, DRPs, or estimated residuals as the response variables for GWAS of longitudinal traits were also applied for extensive comparisons. Details of each model were specified below as well as listed in Table . [...] For each SNP, the incremental Wald statistic implemented by ASReml was used to examine whether the SNP is associated with the trait. The Wald chi-squared statistic with a degree of freedom of df w is given by:W=R(fullmodel)-R(reducedmodel)σ^e2. Here, [R(fullmodel)-R(reducedmodel)] denotes the difference between the reduction in the sums of squares (RSS) or models with and without SNP effect. The symbol df w is degree of freedom for the SNP effect. For fGWAS-C model, df w = nf + 1, and for fGWAS-F model, df w = 2(nf + 1), where nf is the order of basis functions for the time-varied SNP effect as defined above. For other models defined above, df w = 1. The symbol σ^e2 is residual variance estimated via residual maximum likelihood (REML) method. [...] We performed extensive simulations to systematically compare the performance of two random regression-based GWAS models (fGWAS-C and fGWAS-F) proposed here and other multiple-step traditional linear mixed models (GWAS-EBV-P, GWAS-EBV-NP, GWAS-DRP-P, GWAS-DRP-NP and GWAS-Residual) aforementioned. We evaluated statistical power, type-I error rate as well as the accuracy of SNP effect estimated for each GWAS method through 1,000 replication.Population and genomic data were simulated with QMSim software. The simulation started with a base population of 50 males and 50 females in generation −1,000, followed by 1,000 discrete historical generations (generation −1,000 to −1) with the same population size and an equal sex ratio. After 1,000 historical generations, the recent population was generated from generation −1 to generation 0 with population size expanded from 100 to 1,000 (500 males and 500 females). In the follow-up four recent generations (generation 1 to 4), 50 males were randomly selected from the last generation each mating with 500 females. Each female produced two offspring (one male and one female) at each recent generation. Finally, a total of 2,000 females from the last four recent generations were collected as the experimental population investigated. We determined 1,002 SNPs as the simulated genomic data, two of which were selected as independent target mutations. One contributed to the genetic variance (treated as the QTN) and the other had null effect on the longitudinal phenotype. These two SNPs were adopted to evaluate statistical powers and FPRs respectively across different models. The remaining 1000 SNPs were assigned the polygenic effects representing the genetic background of each individual, all genotypes of which were then removed in the final simulated data.The longitudinal phenotype observations were simulated using self-developed C program. The detailed description was given in Supplementary Methods. In the simulation, heritability of simulated trait h 2 was set to 0.3 and heritability of functional QTN hQTN2 (the proportion of phenotypic variance explained by the QTN) was set to different levels of 0.1%, 0.5%, 1% and 2%. The variances explained by the polygenetic and permanent environmental effect were scaled to achieve the preset heritability of the simulated trait.In the simulation, the power and type-I error rate for each scenario were determined as the proportion of significant detections for functional QTN and null-effect SNP respectively among 1,000 replicates for each scenario. […]

Pipeline specifications

Software tools fgwas, ASREML, QMSim
Application GWAS
Organisms Human gammaherpesvirus 4, Bos taurus
Chemicals Nucleotides