Genomic Prediction.md

In allogamous/EnvRtype: Envirotyping Tools for Crop Research and Genetic Improvment

Genomic Prediction using Environmental Covariates and Considering Reaction-Norms

• author: Germano Costa Neto

last update: 13th December 2020

• Software
• Data Sets
• Environmental Covariables (ECs)
• Environmental Relatedness Kernels
• Preparing the Kernels for Prediction
• Fitting bayesian kernel models
• Cross-validation to assess predictive ability
• Codes (G3 paper)

# Software ```{r, eval=FALSE} library(devtools) install_github('allogamous/EnvRtype') library(EnvRtype) wzxhzdk:0

### Environmental Covariables (ECs) for **W** Matrix (W.matrix function) ```{r, eval=FALSE} ## Organizing Environmental Covariables (ECs) in W matrix > * Data were organized for different development stages in maize. We assume fixed time intervals from the days after planting. stages = c('VE','V1_V6','V6_VT','VT_R1','R1_R3','R3_R6',"H") interval = c(0,7,30,65,70,84,105) id.vars = names(maizeWTH)[c(10:15,23,25:30)] W.matrix = W_matrix(env.data = maizeWTH,env.id = 'env', var.id = id.vars,by.interval = T,time.window = interval, names.window = stages,center = F,scale = F ) wzxhzdk:1

### Preparing the Kernels for Prediction (get_kernel function) > * the get_kernel function creates the modeling structure for predictive purposes > * In this example, we show the use of the Reaction-Norm Main Effect Model (RNMM), assuming: y = intercept + fixed effects + enviromic + genomic + enviromic x genomic + error. ```{r, eval=FALSE} ## Assembly Genomic and Enviromic Kernel Models M1 = get_kernel(K_G = K_G,data = maizeYield,env = env,gid = gid,y = y, model = "MDs") # baseline model M2 = get_kernel(K_G = K_G, K_E = K_F, data = maizeYield,env = env,gid = gid,y = y, model = "RNMM",dimension_KE = 'q') # reaction-norm 1 M3 = get_kernel(K_G = K_G, K_E = K_S, data = maizeYield,env = env,gid = gid,y = y,model = "RNMM",dimension_KE = 'q') # reaction-norm 2 wzxhzdk:2

### Cross-validation to assess predictive ability of GP models (kernel_model function) > * creating the training sets (prediction scenario: prediction of novel genotypes, CV1) > * to speed up your analysis, we suggest to use foreach to run the cross-validation ```{r, eval=FALSE} source('https://raw.githubusercontent.com/gcostaneto/SelectivePhenotyping/master/cvrandom.R') rep = 10 seed = 7121 f = 0.80 iter = 5E3 burn = 1E3 thin = 10 TS = Sampling.CV1(gids = Y$gid,f = f,seed = seed,rep = rep,gidlevel = F) wzxhzdk:3

### Our Suggestions > * Although we provide a function to run predictions, the user can also use the kernels created in get_kernel to run analyzes on other packages, such as BGLR or asreml. These packages are interesting because they also allow the modeling of the variance-covariance structure of the residues (errors), in addition to allows the inclusion of factor analytic structures, which can increase the predictive ability of the models; > * When running genomic prediction (GP) for multiple environments, we enviasage that CV schemes such as CV0 (prediction of novel environments) and CV00 (prediction of novel genotypes at novel environments) are indispensable to better assess the potential predictive ability of yours GP models. The power of models with ECs will be better highlighted when prediction scenarios involve a lack of environmental inforamtion. Schemes such as CV1 or CV2, perhpas, will not be enought to discriminate the predictive ability of the non-enviromic and enviromic-based GP models. > * Finally, in addition to the predictive ability assessed at model level, for multi-environment conditons is necessary to assess the predictive ability of the models at each environment. Thus, a better discrimination of the model's potentialities can be achvied. --------------------------------------------------------------------------------------- ## G3 Paper: Codes for Genomic Prediction using Reaction-Norm ### CV1 and CV00 > * CV1 ```{r, eval=FALSE} rep = 30 seed = 1010 f = 0.20 iter = 2E3 burn = 5E2 thin = 10 source('https://raw.githubusercontent.com/gcostaneto/SelectivePhenotyping/master/cvrandom.R') TS = Sampling.CV1(gids = Y$gid,f = f,seed = seed,rep = rep,gidlevel = F) require(foreach) require(doParallel) cl <- makeCluster(3) registerDoParallel(cl) results <-foreach(REP = 1:rep, .combine = "rbind")%:% foreach(MODEL = 1:length(model), .combine = "rbind")%dopar% { yNA <- Y tr <- TS[[REP]] yNA$value[-tr] <- NA Z_E = model.matrix(~0+env,data=yNA) # fixed environmental effects fit <- kernel_model(data = yNA,y = y,env = env,gid = gid, random = Models[[MODEL]],fixed = Z_E, iterations = iter,burnin = burn,thining = thin) df<-data.frame(Model = model[MODEL],rep=REP, rTr=cor(Y$value[tr ], fit$fitted$yHat[tr ],use = 'complete.obs'), rTs=cor(Y$value[-tr], fit$fitted$yHat[-tr],use = 'complete.obs')) cat(paste0(model[MODEL],' ',REP," r = ",round(cor(Y$value[-tr], fit$fitted$yHat[-tr]),3),'\n')) write.table(x = df,file = 'PA_models.txt',sep=',',append = T,row.names=T) output <- data.frame(obs=Y$value,pred=fit$fitted$yHat, gid=Y$gid, env=Y$env, Model = model[MODEL],rep=REP,pop=NA) output$pop[tr ] <- 'training' output$pop[-tr] <- 'testing' return(output) } stopCluster(cl) pa = ddply(results,.(rep,pop,Model),summarise,r = cor(obs,pred)) ddply(pa,.(pop,Model),summarise, pa = round(mean(r),3),sd = round(sd(r),3)) wzxhzdk:4

allogamous/EnvRtype documentation built on Nov. 1, 2024, 3:48 a.m.