Last updated: 2025-07-26
Checks: 6 1
Knit directory:
Machine-learning-e-redes-neurais-artificiais-no-melhoramento-genetico-do-cafeeiro/
This reproducible R Markdown analysis was created with workflowr (version 1.7.1). The Checks tab describes the reproducibility checks that were applied when the results were created. The Past versions tab lists the development history.
The R Markdown file has unstaged changes. To know which version of
the R Markdown file created these results, you’ll want to first commit
it to the Git repo. If you’re still working on the analysis, you can
ignore this warning. When you’re finished, you can run
wflow_publish to commit the R Markdown file and build the
HTML.
Great job! The global environment was empty. Objects defined in the global environment can affect the analysis in your R Markdown file in unknown ways. For reproduciblity it’s best to always run the code in an empty environment.
The command set.seed(20250709) was run prior to running
the code in the R Markdown file. Setting a seed ensures that any results
that rely on randomness, e.g. subsampling or permutations, are
reproducible.
Great job! Recording the operating system, R version, and package versions is critical for reproducibility.
Nice! There were no cached chunks for this analysis, so you can be confident that you successfully produced the results during this run.
Great job! Using relative paths to the files within your workflowr project makes it easier to run your code on other machines.
Great! You are using Git for version control. Tracking code development and connecting the code version to the results is critical for reproducibility.
The results in this page were generated with repository version 82b3f68. See the Past versions tab to see a history of the changes made to the R Markdown and HTML files.
Note that you need to be careful to ensure that all relevant files for
the analysis have been committed to Git prior to generating the results
(you can use wflow_publish or
wflow_git_commit). workflowr only checks the R Markdown
file, but you know if there are other scripts or data files that it
depends on. Below is the status of the Git repository when the results
were generated:
Ignored files:
Ignored: .Rproj.user/
Ignored: analysis/modelos_mistos_dados_reais.Rmd
Ignored: data/dados_reais/Fenótipos 2014_2015_2016_C.arabica (1).xlsx
Ignored: data/dados_reais/Fenótipos 2014_2015_2016_C.arabica.xlsx
Ignored: data/dados_reais/FiltStep1_minQ10.0_minDP3_DPrange15-maxMeanDepth750_miss0.4_maf0.01_mac1_EMB_291701_SNPs_Raw.vcf.012.012
Ignored: data/dados_reais/FiltStep1_minQ10.0_minDP3_DPrange15-maxMeanDepth750_miss0.4_maf0.01_mac1_EMB_291701_SNPs_Raw.vcf.012.012.indv
Ignored: data/dados_reais/FiltStep1_minQ10.0_minDP3_DPrange15-maxMeanDepth750_miss0.4_maf0.01_mac1_EMB_291701_SNPs_Raw.vcf.012.012.pos
Ignored: data/dados_reais/FiltStep1_minQ10_minDP3_DPrange15-750_miss0.4_maf0.01_mac1_EMB_291701_filt1_snps_RAW.012.indv.txt
Ignored: data/dados_reais/FiltStep1_minQ10_minDP3_DPrange15-750_miss0.4_maf0.01_mac1_EMB_291701_filt1_snps_RAW.012.pos.txt
Ignored: data/dados_reais/FiltStep1_minQ10_minDP3_DPrange15-750_miss0.4_maf0.01_mac1_EMB_291701_filt1_snps_RAW.012.txt
Ignored: data/dados_reais/fpls-09-01934.pdf
Ignored: data/dados_reais/plants-13-01876-v2.pdf
Ignored: output/BLUPS_par_mmer.Rdata
Ignored: output/blups_all_wide.csv
Ignored: output/mean_pheno.csv
Ignored: output/real_results_consolidated_10r_3f.xlsx
Ignored: output/real_results_consolidated_5r_5f.xlsx
Ignored: output/result_sommer_pl_geracao_random.RDS
Ignored: output/result_sommer_random.RDS
Untracked files:
Untracked: data/dados_reais/geno_real.rds
Untracked: data/dados_reais/pheno_real.rds
Untracked: data/dados_reais/pos_geno_real.rds
Untracked: output/gwas_combined_results.rds
Untracked: output/gwas_results.rds
Untracked: output/importance_ML.rds
Untracked: output/results_cv_GBLUP_a.rds
Untracked: output/results_cv_GBLUP_ad.rds
Untracked: output/results_cv_GBLUP_ade.rds
Untracked: output/simulated_results_consolidated.xlsx
Unstaged changes:
Modified: analysis/_site.yml
Modified: analysis/gwas_dados_reais.Rmd
Modified: analysis/gwas_dados_simulados.Rmd
Modified: analysis/gws_dados_reais.Rmd
Modified: analysis/gws_dados_simulados.Rmd
Note that any generated files, e.g. HTML, png, CSS, etc., are not included in this status report because it is ok for generated content to have uncommitted changes.
These are the previous versions of the repository in which changes were
made to the R Markdown (analysis/gws_dados_simulados.Rmd)
and HTML (docs/gws_dados_simulados.html) files. If you’ve
configured a remote Git repository (see ?wflow_git_remote),
click on the hyperlinks in the table below to view the files as they
were in that past version.
| File | Version | Author | Date | Message |
|---|---|---|---|---|
| Rmd | 60e67cf | WevertonGomesCosta | 2025-07-24 | chance dados_simulados.Rmd to gws ans gwas_dados_simulados.Rmd |
nfolds <- 5
nrept <- 5
traits <- colnames(pheno)[-1]
set.seed(123)
# Alinhamento de IDs
pheno <- pheno[order(pheno$ID), ]
geno <- geno[order(rownames(geno)), ]
stopifnot(all(rownames(geno) == as.character(pheno$ID)))
geno <- geno - 1
# Generating fold list
fold.list <- vector("list", length = nrept)
for (r in seq_len(nrept)) {
folds <- cvFolds(
n = nrow(pheno),
K = nfolds,
R = r,
type = "random"
)
df <- cbind(folds$which, folds$subsets)
fold.list[[r]] <- split(df[, 2], df[, 1])
}O modelo G-BLUP (Genomic Best Linear Unbiased Prediction) é uma abordagem amplamente utilizada para predição genômica, que considera tanto os efeitos aditivos quanto os efeitos de dominância dos marcadores genéticos. Neste estudo, implementamos três variantes do G-BLUP: Aditivo, Aditivo-Dominante e Aditivo-Dominante-Epistástico.
O modelo é implementado utilizando a função mmes do
pacote sommer, que permite ajustar modelos mistos com
estruturas de parentesco complexas. A validação cruzada é realizada para
avaliar a acurácia preditiva do modelo.
Para o modelo G-BLUP aditivo, utilizamos a matriz de parentesco aditivo (K) calculada a partir dos dados genotípicos. Este modelo assume que os efeitos dos marcadores são puramente aditivos, sem considerar interações de dominância ou epistasia.
cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_GBLUP_a <- foreach(
idx = seq_len(nrow(combos)),
.packages = c("sommer", "dplyr", "tibble"),
.combine = bind_rows
) %dopar% {
# Extração dos parâmetros do grid
tr <- combos$trait[idx]
r <- combos$rep[idx]
f <- combos$fold[idx]
# Seleção dos dados fenotípicos e matrizes de parentesco
ph <- pheno[,c("ID", all_of(tr))]
K1 <- K[ph$ID, ph$ID]
td <- ph
# Definição dos índices de validação cruzada
val_idx <- fold.list[[r]][[f]]
td[val_idx, tr] <- NA
# Ajuste do modelo G-BLUP aditivo
fit <- mmes(
fixed = as.formula(paste(tr, "~ 1")),
random = ~ vsm(ism(ID), Gu = K1),
data = td,
rcov = ~ units
)
# Predição e métricas
pred <- data.frame(yhat = fit$u + fit$b[[1]], ph)
# Cálculo dos GEBVs para treino e validação
pred_T <- pred[!(pred$ID %in% val_idx), "yhat"]
truth_T <- pred[[tr]][!(pred$ID %in% val_idx)]
pred_V <- pred[pred$ID %in% val_idx, "yhat"]
truth_V <- pred[[tr]][pred$ID %in% val_idx]
# Criação do tibble com os resultados
tibble(
Trait = tr,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, truth_T), 4),
R2_Valid = round(cor(pred_V, truth_V), 4),
RMSE_Train = round(sqrt(mean((pred_T - truth_T)^2)), 4),
RMSE_Valid = round(sqrt(mean((pred_V - truth_V)^2)), 4),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V)
)
}
stopCluster(cl)
saveRDS(results_cv_GBLUP_a, file = "output/results_cv_GBLUP_a.rds")Para o modelo G-BLUP aditivo-dominante, utilizamos a matriz de parentesco aditivo-dominante (K) que considera tanto os efeitos aditivos quanto os de dominância dos marcadores genéticos. Este modelo é mais complexo e pode capturar interações entre alelos.
cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_ad <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_GBLUP_ad <- foreach(
idx = seq_len(nrow(combos_ad)),
.packages = c("sommer", "dplyr", "tibble"),
.combine = bind_rows
) %dopar% {
idx=1
# Extração dos parâmetros do grid
tr <- combos_ad$trait[idx]
r <- combos_ad$rep[idx]
f <- combos_ad$fold[idx]
# Seleção dos dados fenotípicos e matrizes de parentesco
ph <- pheno[,c("ID", all_of(tr))]
K1 <- K[ph$ID, ph$ID]
D1 <- D[ph$ID, ph$ID]
# Criação dos IDs para os efeitos aditivo e dominante
td <- ph
td$IDD <- td$ID
# Definição dos índices de validação cruzada
val_idx <- fold.list[[r]][[f]]
td[val_idx, tr] <- NA
# Ajuste do modelo G-BLUP aditivo-dominante
fit <- mmes(
fixed = as.formula(paste(tr, "~ 1")),
random = ~ vsm(ism(ID), Gu = K1) + vsm(ism(IDD), Gu = D1),
data = td,
rcov = ~ units
)
# Extração dos preditos e verdadeiros valores
pred_a <- fit$uList$`vsm(ism(ID), Gu = K1`
pred_d <- fit$uList$`vsm(ism(IDD), Gu = D1`
# Soma dos preditos aditivos e dominantes
pred <- pred_a + pred_d + fit$b[[1]]
truth <- ph[[tr]]
# Predição e métricas
pred <- data.frame(pred, ph)
# Cálculo dos GEBVs para treino e validação
pred_T <- pred[-val_idx, "mu"]
truth_T <- pred[[tr]][-val_idx]
pred_V <- pred[val_idx, "mu"]
truth_V <- pred[[tr]][val_idx]
# Criação do tibble com os resultados
tibble(
Trait = tr,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, truth_T), 4),
R2_Valid = round(cor(pred_V, truth_V), 4),
RMSE_Train = round(sqrt(mean((pred_T - truth_T)^2)), 4),
RMSE_Valid = round(sqrt(mean((pred_V - truth_V)^2)), 4),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V)
)
}
stopCluster(cl)
saveRDS(results_cv_GBLUP_ad, file = "output/results_cv_GBLUP_ad.rds")Para o modelo G-BLUP aditivo-dominante-epistástico, utilizamos a matriz de parentesco que inclui efeitos epistáticos. Este modelo é o mais complexo e pode capturar interações entre múltiplos marcadores.
cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_ade <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_GBLUP_ade <- foreach(
idx = seq_len(nrow(combos_ade)),
.packages = c("sommer", "dplyr", "tibble"),
.combine = bind_rows
) %dopar% {
# Extração dos parâmetros do grid
tr <- combos_ade$trait[idx]
r <- combos_ade$rep[idx]
f <- combos_ade$fold[idx]
# Seleção dos dados fenotípicos e matrizes de parentesco
ph <- pheno[,c("ID", all_of(tr))]
K1 <- K[ph$ID, ph$ID]
D1 <- D[ph$ID, ph$ID]
E1 <- E[ph$ID, ph$ID]
# Criação dos IDs para os efeitos aditivo, dominante e epistático
td <- ph
td$IDD <- td$ID
td$IDE <- td$ID
# Definição dos índices de validação cruzada
val_idx <- fold.list[[r]][[f]]
td[val_idx, tr] <- NA
# Ajuste do modelo G-BLUP aditivo-dominante-epistástico
fit <- mmes(
fixed = as.formula(paste(tr, "~ 1")),
random = ~ vsm(ism(ID), Gu = K1) + vsm(ism(IDD), Gu = D1) + vsm(ism(IDE), Gu = E1),
data = td,
rcov = ~ units
)
# Extração dos preditos e verdadeiros valores
pred_a <- fit$uList$`vsm(ism(ID), Gu = K1`
pred_d <- fit$uList$`vsm(ism(IDD), Gu = D1`
pred_e <- fit$uList$`vsm(ism(IDE), Gu = E1`
# Soma dos preditos aditivos e dominantes
pred <- pred_a + pred_d + pred_e + fit$b[[1]]
truth <- ph[[tr]]
# Predição e métricas
pred <- data.frame(pred, ph)
# Cálculo dos GEBVs para treino e validação
pred_T <- pred[-val_idx, "mu"]
truth_T <- pred[[tr]][-val_idx]
pred_V <- pred[val_idx, "mu"]
truth_V <- pred[[tr]][val_idx]
# Criação do tibble com os resultados
tibble(
Trait = tr,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, truth_T), 4),
R2_Valid = round(cor(pred_V, truth_V), 4),
RMSE_Train = round(sqrt(mean((pred_T - truth_T)^2)), 4),
RMSE_Valid = round(sqrt(mean((pred_V - truth_V)^2)), 4),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V)
)
}
stopCluster(cl)
saveRDS(results_cv_GBLUP_ade, file = "output/results_cv_GBLUP_ade.rds")cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_mars <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_MARS_1 <- foreach(
idx = seq_len(nrow(combos_mars)),
.packages = c("earth", "dplyr", "tibble"),
.combine = bind_rows
) %dopar% {
tr <- combos_mars$trait[idx]
r <- combos_mars$rep[idx]
f <- combos_mars$fold[idx]
valid_idx <- fold.list[[r]][[f]]
train_idx <- setdiff(seq_len(nrow(pheno)), valid_idx)
yT <- pheno[train_idx, tr]
xT <- geno[train_idx, ]
yV <- pheno[valid_idx, tr]
xV <- geno[valid_idx, ]
fit <- earth(x = xT, y = yT, degree = 1)
pred_T <- predict(fit, xT)
pred_V <- predict(fit, xV)
# Extrair importância das variáveis
imp <- evimp(fit, trim = FALSE)
# Seleciona colunas relevantes e converte para data.frame
imp_df <- as.data.frame(imp[, c("rss", "nsubsets")]) # rss: Overall importance
# Adiciona informações de repetição e fold, mantendo nomes dos marcadores
imp_df$marker <- rownames(imp_df)
imp_df$Rep <- r
imp_df$Fold <- f
# Ajusta nomes das colunas e remove sufixo indesejado dos marcadores
imp_df <- imp_df %>%
select(Overall = rss, marker, Rep, Fold) %>%
mutate(marker = stringr::str_replace_all(marker, "-unused", ""))
gc()
tibble(
Trait = tr,
Degree = 1,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, yT), 4),
R2_Valid = round(cor(pred_V, yV), 4),
RMSE_Train = round(sqrt(mean((pred_T - yT)^2)), 4),
RMSE_Valid = round(sqrt(mean((pred_V - yV)^2)), 4),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V),
imp = list(imp_df)
)
}
stopCluster(cl)
saveRDS(results_cv_MARS_1, file = "output/results_cv_MARS_1.rds")cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_mars <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_MARS_2 <- foreach(
idx = seq_len(nrow(combos_mars)),
.packages = c("earth", "dplyr", "tibble"),
.combine = bind_rows
) %dopar% {
tr <- combos_mars$trait[idx]
r <- combos_mars$rep[idx]
f <- combos_mars$fold[idx]
valid_idx <- fold.list[[r]][[f]]
train_idx <- setdiff(seq_len(nrow(pheno)), valid_idx)
yT <- pheno[train_idx, tr]
xT <- geno[train_idx, ]
yV <- pheno[valid_idx, tr]
xV <- geno[valid_idx, ]
fit <- earth(x = xT, y = yT, degree = 2)
pred_T <- predict(fit, xT)
pred_V <- predict(fit, xV)
# Extrair importância das variáveis
imp <- evimp(fit, trim = FALSE)
# Seleciona colunas relevantes e converte para data.frame
imp_df <- as.data.frame(imp[, c("rss", "nsubsets")]) # rss: Overall importance
# Adiciona informações de repetição e fold, mantendo nomes dos marcadores
imp_df$marker <- rownames(imp_df)
imp_df$Rep <- r
imp_df$Fold <- f
# Ajusta nomes das colunas e remove sufixo indesejado dos marcadores
imp_df <- imp_df %>%
select(Overall = rss, marker, Rep, Fold) %>%
mutate(marker = stringr::str_replace_all(marker, "-unused", ""))
gc()
tibble(
Trait = tr,
Degree = 2,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, yT), 4),
R2_Valid = round(cor(pred_V, yV), 4),
RMSE_Train = round(sqrt(mean((pred_T - yT)^2)), 4),
RMSE_Valid = round(sqrt(mean((pred_V - yV)^2)), 4),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V),
imp = list(imp_df)
)
}
stopCluster(cl)
saveRDS(results_cv_MARS_2, file = "output/results_cv_MARS_2.rds")cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_mars <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_MARS_3 <- foreach(
idx = seq_len(nrow(combos_mars)),
.packages = c("earth", "dplyr", "tibble"),
.combine = bind_rows
) %dopar% {
tr <- combos_mars$trait[idx]
r <- combos_mars$rep[idx]
f <- combos_mars$fold[idx]
valid_idx <- fold.list[[r]][[f]]
train_idx <- setdiff(seq_len(nrow(pheno)), valid_idx)
yT <- pheno[train_idx, tr]
xT <- geno[train_idx, ]
yV <- pheno[valid_idx, tr]
xV <- geno[valid_idx, ]
fit <- earth(x = xT, y = yT, degree = 3)
pred_T <- predict(fit, xT)
pred_V <- predict(fit, xV)
# Extrair importância das variáveis
imp <- evimp(fit, trim = FALSE)
# Seleciona colunas relevantes e converte para data.frame
imp_df <- as.data.frame(imp[, c("rss", "nsubsets")]) # rss: Overall importance
# Adiciona informações de repetição e fold, mantendo nomes dos marcadores
imp_df$marker <- rownames(imp_df)
imp_df$Rep <- r
imp_df$Fold <- f
# Ajusta nomes das colunas e remove sufixo indesejado dos marcadores
imp_df <- imp_df %>%
select(Overall = rss, marker, Rep, Fold) %>%
mutate(marker = stringr::str_replace_all(marker, "-unused", ""))
gc()
tibble(
Trait = tr,
Degree = 3,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, yT), 4),
R2_Valid = round(cor(pred_V, yV), 4),
RMSE_Train = round(sqrt(mean((pred_T - yT)^2)), 4),
RMSE_Valid = round(sqrt(mean((pred_V - yV)^2)), 4),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V),
imp = list(imp_df)
)
}
stopCluster(cl)
saveRDS(results_cv_MARS_3, file = "output/results_cv_MARS_3.rds")cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_dt <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_DT <- foreach(
idx = seq_len(nrow(combos_dt)),
.packages = c("rpart", "dplyr", "tibble", "caret"),
.combine = bind_rows
) %dopar% {
tr <- combos_dt$trait[idx]
r <- combos_dt$rep[idx]
f <- combos_dt$fold[idx]
valid_idx <- fold.list[[r]][[f]]
train_idx <- setdiff(seq_len(nrow(pheno)), valid_idx)
yT <- pheno[[tr]][train_idx] # vetor
xT <- data.frame(geno[train_idx, ])
yV <- pheno[[tr]][valid_idx]
xV <- data.frame(geno[valid_idx, ])
fit <- rpart(yT ~ ., data = xT)
pred_T <- predict(fit, xT)
pred_V <- predict(fit, xV)
imp_df <- data.frame(varImp(fit, scale = TRUE, value = "rss"))
# Adiciona informações de repetição e fold, mantendo nomes dos marcadores
imp_df$marker <- rownames(imp_df)
imp_df$Rep <- r
imp_df$Fold <- f
gc()
tibble(
Trait = tr,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, yT), 3),
R2_Valid = round(cor(pred_V, yV), 3),
RMSE_Train = round(sqrt(mean((pred_T - yT)^2)), 3),
RMSE_Valid = round(sqrt(mean((pred_V - yV)^2)), 3),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V),
imp = list(imp_df)
)
}
stopCluster(cl)
saveRDS(results_cv_DT, file = "output/results_cv_DT.rds")cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_bag <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_Bag <- foreach(
idx = seq_len(nrow(combos_bag)),
.packages = c("randomForest", "dplyr", "tibble", "caret"),
.combine = bind_rows
) %dopar% {
tr <- combos_bag$trait[idx]
r <- combos_bag$rep[idx]
f <- combos_bag$fold[idx]
valid_idx <- fold.list[[r]][[f]]
train_idx <- setdiff(seq_len(nrow(pheno)), valid_idx)
yT <- pheno[[tr]][train_idx] # vetor
xT <- geno[train_idx, ]
yV <- pheno[[tr]][valid_idx]
xV <- geno[valid_idx, ]
fit <- randomForest(x = xT, y = yT, mtry = ncol(xT))
pred_T <- predict(fit, xT)
pred_V <- predict(fit, xV)
imp_df <- data.frame(varImp(fit, scale = TRUE, value = "rss"))
# Adiciona informações de repetição e fold, mantendo nomes dos marcadores
imp_df$marker <- rownames(imp_df)
imp_df$Rep <- r
imp_df$Fold <- f
gc()
tibble(
Trait = tr,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, yT), 3),
R2_Valid = round(cor(pred_V, yV), 3),
RMSE_Train = round(sqrt(mean((pred_T - yT)^2)), 3),
RMSE_Valid = round(sqrt(mean((pred_V - yV)^2)), 3),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V),
imp = list(imp_df)
)
}
stopCluster(cl)
saveRDS(results_cv_Bag, file = "output/results_cv_Bag.rds")cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_rf <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_RF <- foreach(
idx = seq_len(nrow(combos_rf)),
.packages = c("randomForest", "dplyr", "tibble", "caret"),
.combine = bind_rows
) %dopar% {
tr <- combos_rf$trait[idx]
r <- combos_rf$rep[idx]
f <- combos_rf$fold[idx]
valid_idx <- fold.list[[r]][[f]]
train_idx <- setdiff(seq_len(nrow(pheno)), valid_idx)
yT <- pheno[[tr]][train_idx] # vetor
xT <- geno[train_idx, ]
yV <- pheno[[tr]][valid_idx]
xV <- geno[valid_idx, ]
fit <- randomForest(xT, yT, mtry = floor(ncol(xT) / 3), ntree = 500)
pred_T <- predict(fit, xT)
pred_V <- predict(fit, xV)
imp_df <- data.frame(varImp(fit, scale = TRUE, value = "rss"))
# Adiciona informações de repetição e fold, mantendo nomes dos marcadores
imp_df$marker <- rownames(imp_df)
imp_df$Rep <- r
imp_df$Fold <- f
gc()
tibble(
Trait = tr,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, yT), 3),
R2_Valid = round(cor(pred_V, yV), 3),
RMSE_Train = round(sqrt(mean((pred_T - yT)^2)), 3),
RMSE_Valid = round(sqrt(mean((pred_V - yV)^2)), 3),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V),
imp = list(imp_df)
)
}
stopCluster(cl)
saveRDS(results_cv_RF, file = "output/results_cv_RF.rds")cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_gbm <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
stringsAsFactors = FALSE
)
results_cv_GBM <- foreach(
idx = seq_len(nrow(combos_gbm)),
.packages = c("gbm", "dplyr", "tibble", "caret"),
.combine = bind_rows
) %dopar% {
tr <- combos_gbm$trait[idx]
r <- combos_gbm$rep[idx]
f <- combos_gbm$fold[idx]
valid_idx <- fold.list[[r]][[f]]
train_idx <- setdiff(seq_len(nrow(pheno)), valid_idx)
yT <- pheno[[tr]][train_idx] # vetor
xT <- geno[train_idx, ]
yV <- pheno[[tr]][valid_idx]
xV <- geno[valid_idx, ]
fit <- gbm(
yT ~ .,
data = data.frame(yT, xT),
distribution = "gaussian",
n.trees = 500,
interaction.depth = 2
)
pred_T <- predict(fit, data.frame(xT), n.trees = 500)
pred_V <- predict(fit, data.frame(xV), n.trees = 500)
imp_df <- data.frame(varImp(fit, scale = TRUE, value = "rss" , numTrees = 500))
# Adiciona informações de repetição e fold, mantendo nomes dos marcadores
imp_df$marker <- rownames(imp_df)
imp_df$Rep <- r
imp_df$Fold <- f
gc()
tibble(
Trait = tr,
Rep = r,
Fold = f,
R2_Train = round(cor(pred_T, yT), 3),
R2_Valid = round(cor(pred_V, yV), 3),
RMSE_Train = round(sqrt(mean((pred_T - yT)^2)), 3),
RMSE_Valid = round(sqrt(mean((pred_V - yV)^2)), 3),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V),
imp = list(imp_df)
)
}
stopCluster(cl)
saveRDS(results_cv_GBM, file = "output/results_cv_GBM.rds")cl <- makeCluster(max(1, detectCores() - 1))
registerDoParallel(cl)
combos_nn <- expand.grid(
trait = traits,
rep = seq_len(nrept),
fold = seq_len(nfolds),
h1 = c(1, 2, 4, 8, 16, 32),
h2 = c(0, 1, 2, 4, 8, 16, 32),
stringsAsFactors = FALSE
)
results_cv_NN <- foreach(
idx = seq_len(nrow(combos_nn)),
.packages = c("ANN2", "neuralnet", "dplyr", "tibble"),
.combine = bind_rows
) %dopar% {
tr <- combos_nn$trait[idx]
r <- combos_nn$rep[idx]
f <- combos_nn$fold[idx]
h1 <- combos_nn$h1[idx]
h2 <- combos_nn$h2[idx]
hidden.layers <- case_when(h2 != 0 ~ c(h1, h2),
TRUE ~ h1)
valid_idx <- fold.list[[r]][[f]]
train_idx <- setdiff(seq_len(nrow(pheno)), valid_idx)
yT <- pheno[[tr]][train_idx] # vetor
xT <- geno[train_idx, ]
yV <- pheno[[tr]][valid_idx]
xV <- geno[valid_idx, ]
# Construção e Treinamento da Rede Neural
NN <- neuralnetwork(
X = xT,
y = yT,
hidden.layers = hidden.layers,
regression = TRUE,
activ.functions = "relu",
n.epochs = 500,
loss.type = "squared",
val.prop = 0
)
pred_T <- predict(NN, xT)$predictions
pred_V <- predict(NN, xV)$predictions
tibble(
Trait = tr,
Rep = r,
Fold = f,
Hidden_Layers_1 = h1,
Hidden_Layers_2 = h2,
R2_Train = round(cor(pred_T, yT), 3),
R2_Valid = round(cor(pred_V, yV), 3),
RMSE_Train = round(sqrt(mean((pred_T - yT)^2)), 3),
RMSE_Valid = round(sqrt(mean((pred_V - yV)^2)), 3),
GEBV_T = list(pred_T),
GEBV_V = list(pred_V)
)
}
stopCluster(cl)
saveRDS(results_cv_NN, file = "output/results_cv_NN.rds")# Carregar resultados
results_cv_GBLUP_a <- readRDS("output/results_cv_GBLUP_a.rds")
results_cv_GBLUP_ad <- readRDS("output/results_cv_GBLUP_ad.rds")
results_cv_GBLUP_ade <- readRDS("output/results_cv_GBLUP_ade.rds")
results_cv_MARS_1 <- readRDS("output/results_cv_MARS_1.rds")
results_cv_MARS_2 <- readRDS("output/results_cv_MARS_2.rds")
results_cv_MARS_3 <- readRDS("output/results_cv_MARS_3.rds")
results_cv_DT <- readRDS("output/results_cv_DT.rds")
results_cv_Bag <- readRDS("output/results_cv_Bag.rds")
results_cv_RF <- readRDS("output/results_cv_RF.rds")
results_cv_GBM <- readRDS("output/results_cv_GBM.rds")
results_cv_NN <- readRDS("output/results_cv_NN.rds")
# Agrupa por Trait e configuração de neurônios
resumo_nn <- results_cv_NN %>%
group_by(Trait, Hidden_Layers_1, Hidden_Layers_2) %>%
summarise(
R2_Valid_Mean = mean(R2_Valid, na.rm = TRUE),
R2_Train_Mean = mean(R2_Train, na.rm = TRUE),
RMSE_Valid_Mean = mean(RMSE_Valid, na.rm = TRUE),
RMSE_Train_Mean = mean(RMSE_Train, na.rm = TRUE),
R2_Valid_SD = sd(R2_Valid, na.rm = TRUE),
R2_Train_SD = sd(R2_Train, na.rm = TRUE),
RMSE_Valid_SD = sd(RMSE_Valid, na.rm = TRUE),
RMSE_Train_SD = sd(RMSE_Train, na.rm = TRUE),
.groups = "drop"
)
# Exibe os melhores por característica com base no maior R²_Valid
melhores_config_por_trait <- resumo_nn %>%
group_by(Trait) %>%
slice_max(order_by = R2_Valid_Mean, n = 1) %>%
arrange(Trait)
print(melhores_config_por_trait)# A tibble: 6 × 11
# Groups: Trait [6]
Trait Hidden_Layers_1 Hidden_Layers_2 R2_Valid_Mean R2_Train_Mean
<chr> <dbl> <dbl> <dbl> <dbl>
1 V1 32 32 0.438 1
2 V2 32 32 0.537 1
3 V3 32 2 0.457 0.951
4 V4 32 0 0.444 1
5 V5 1 2 0.322 0.801
6 V6 16 2 0.306 0.933
# ℹ 6 more variables: RMSE_Valid_Mean <dbl>, RMSE_Train_Mean <dbl>,
# R2_Valid_SD <dbl>, R2_Train_SD <dbl>, RMSE_Valid_SD <dbl>,
# RMSE_Train_SD <dbl># Adicionar coluna Method e combinar todos
combined_results <- bind_rows(
results_cv_GBLUP_a %>% mutate(Method = "GBLUP_a"),
results_cv_GBLUP_ad %>% mutate(Method = "GBLUP_ad"),
results_cv_GBLUP_ade %>% mutate(Method = "GBLUP_ade"),
results_cv_MARS_1 %>% mutate(Method = "MARS_1"),
results_cv_MARS_2 %>% mutate(Method = "MARS_2"),
results_cv_MARS_3 %>% mutate(Method = "MARS_3"),
results_cv_DT %>% mutate(Method = "DT"),
results_cv_Bag %>% mutate(Method = "BAG"),
results_cv_RF %>% mutate(Method = "RF"),
results_cv_GBM %>% mutate(Method = "GBM")
)
# Resumir por Method e Trait
table_final <- combined_results %>%
group_by(Method, Trait) %>%
summarise(
R2_Valid_Mean = mean(R2_Valid, na.rm = TRUE),
R2_Valid_SD = sd(R2_Valid, na.rm = TRUE),
RMSE_Valid_Mean = mean(RMSE_Valid, na.rm = TRUE),
RMSE_Valid_SD = sd(RMSE_Valid, na.rm = TRUE)
) %>%
ungroup() %>%
bind_rows(
melhores_config_por_trait %>%
mutate(Method = "NN") %>%
dplyr::select(Method, Trait, R2_Valid_Mean, R2_Valid_SD, RMSE_Valid_Mean, RMSE_Valid_SD)
)`summarise()` has grouped output by 'Method'. You can override using the
`.groups` argument.# Exibir tabela consolidada
table_final %>%
arrange(Trait, Method) %>%
mutate_if(is.numeric, round, 4) %>%
kable(booktabs = TRUE, caption = "Resultados Consolidados por Método e Traço") %>%
kable_styling(full_width = FALSE)| Method | Trait | R2_Valid_Mean | R2_Valid_SD | RMSE_Valid_Mean | RMSE_Valid_SD |
|---|---|---|---|---|---|
| BAG | V1 | 0.7437 | 0.0354 | 1.5453 | 0.0972 |
| DT | V1 | 0.6573 | 0.0435 | 1.7373 | 0.0938 |
| GBLUP_a | V1 | 0.6401 | 0.0337 | 1.7784 | 0.0752 |
| GBLUP_ad | V1 | 0.6327 | 0.0000 | 1.8537 | 0.0000 |
| GBLUP_ade | V1 | 0.6118 | 0.0289 | 1.8123 | 0.0798 |
| GBM | V1 | 0.7209 | 0.0230 | 1.5889 | 0.0641 |
| MARS_1 | V1 | 0.6571 | 0.0290 | 1.7327 | 0.0746 |
| MARS_2 | V1 | 0.7653 | 0.0202 | 1.4861 | 0.0642 |
| MARS_3 | V1 | 0.6712 | 0.0314 | 1.8404 | 0.1095 |
| NN | V1 | 0.4380 | 0.0441 | 2.2661 | 0.1023 |
| RF | V1 | 0.7444 | 0.0346 | 1.5447 | 0.0953 |
| BAG | V2 | 0.6656 | 0.0345 | 0.8482 | 0.0464 |
| DT | V2 | 0.3090 | 0.0465 | 1.0910 | 0.0494 |
| GBLUP_a | V2 | 0.7164 | 0.0445 | 0.7513 | 0.0485 |
| GBLUP_ade | V2 | 0.6716 | 0.0269 | 0.7966 | 0.0376 |
| GBM | V2 | 0.6910 | 0.0268 | 0.7790 | 0.0431 |
| MARS_1 | V2 | 0.4746 | 0.0456 | 0.9599 | 0.0464 |
| MARS_2 | V2 | 0.5368 | 0.0529 | 0.9383 | 0.0573 |
| MARS_3 | V2 | 0.4912 | 0.0550 | 1.0852 | 0.0867 |
| NN | V2 | 0.5372 | 0.0422 | 0.9836 | 0.0422 |
| RF | V2 | 0.6672 | 0.0329 | 0.8480 | 0.0485 |
| BAG | V3 | 0.5750 | 0.0504 | 0.7896 | 0.0382 |
| DT | V3 | 0.2526 | 0.0653 | 0.9650 | 0.0344 |
| GBLUP_a | V3 | 0.6668 | 0.0284 | 0.6989 | 0.0269 |
| GBLUP_ade | V3 | 0.5966 | 0.0484 | 0.7508 | 0.0314 |
| GBM | V3 | 0.5481 | 0.0527 | 0.7924 | 0.0294 |
| MARS_1 | V3 | 0.4095 | 0.0595 | 0.8683 | 0.0348 |
| MARS_2 | V3 | 0.4151 | 0.0640 | 0.8962 | 0.0443 |
| MARS_3 | V3 | 0.3782 | 0.0599 | 1.0304 | 0.0640 |
| NN | V3 | 0.4574 | 0.0614 | 0.9127 | 0.0534 |
| RF | V3 | 0.5789 | 0.0525 | 0.7879 | 0.0376 |
| BAG | V4 | 0.5753 | 0.0445 | 0.7185 | 0.0253 |
| DT | V4 | 0.2824 | 0.0650 | 0.8685 | 0.0376 |
| GBLUP_a | V4 | 0.6588 | 0.0439 | 0.6422 | 0.0291 |
| GBLUP_ade | V4 | 0.5962 | 0.0318 | 0.6854 | 0.0238 |
| GBM | V4 | 0.5524 | 0.0307 | 0.7214 | 0.0237 |
| MARS_1 | V4 | 0.4330 | 0.0422 | 0.7812 | 0.0265 |
| MARS_2 | V4 | 0.4319 | 0.0592 | 0.8072 | 0.0326 |
| MARS_3 | V4 | 0.3684 | 0.0523 | 0.9553 | 0.0462 |
| NN | V4 | 0.4437 | 0.0462 | 0.8460 | 0.0346 |
| RF | V4 | 0.5789 | 0.0398 | 0.7178 | 0.0259 |
| BAG | V5 | 0.4299 | 0.0400 | 0.7695 | 0.0352 |
| DT | V5 | 0.1408 | 0.0586 | 0.9162 | 0.0427 |
| GBLUP_a | V5 | 0.5617 | 0.0335 | 0.7021 | 0.0312 |
| GBLUP_ade | V5 | 0.4657 | 0.0358 | 0.7491 | 0.0356 |
| GBM | V5 | 0.3892 | 0.0548 | 0.8036 | 0.0419 |
| MARS_1 | V5 | 0.3223 | 0.0426 | 0.8182 | 0.0360 |
| MARS_2 | V5 | 0.2701 | 0.0714 | 0.8772 | 0.0564 |
| MARS_3 | V5 | 0.2433 | 0.0738 | 1.0260 | 0.0702 |
| NN | V5 | 0.3215 | 0.0676 | 0.8936 | 0.0624 |
| RF | V5 | 0.4356 | 0.0367 | 0.7677 | 0.0346 |
| BAG | V6 | 0.4604 | 0.0395 | 0.6582 | 0.0206 |
| DT | V6 | 0.2557 | 0.0619 | 0.7519 | 0.0285 |
| GBLUP_a | V6 | 0.5565 | 0.0382 | 0.6125 | 0.0232 |
| GBLUP_ade | V6 | 0.4741 | 0.0354 | 0.6484 | 0.0192 |
| GBM | V6 | 0.4062 | 0.0334 | 0.6927 | 0.0205 |
| MARS_1 | V6 | 0.3615 | 0.0443 | 0.6994 | 0.0230 |
| MARS_2 | V6 | 0.3334 | 0.0544 | 0.7352 | 0.0266 |
| MARS_3 | V6 | 0.2519 | 0.0697 | 0.8841 | 0.0508 |
| NN | V6 | 0.3064 | 0.0698 | 0.8151 | 0.0481 |
| RF | V6 | 0.4640 | 0.0355 | 0.6572 | 0.0191 |
# Salvar tabela consolidada
writexl::write_xlsx(table_final, "output/simulated_results_consolidated.xlsx")
R version 4.4.1 (2024-06-14 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 26100)
Matrix products: default
locale:
[1] LC_COLLATE=Portuguese_Brazil.utf8 LC_CTYPE=Portuguese_Brazil.utf8
[3] LC_MONETARY=Portuguese_Brazil.utf8 LC_NUMERIC=C
[5] LC_TIME=Portuguese_Brazil.utf8
time zone: America/Sao_Paulo
tzcode source: internal
attached base packages:
[1] parallel stats graphics grDevices utils datasets methods
[8] base
other attached packages:
[1] caret_7.0-1 lubridate_1.9.4 forcats_1.0.0
[4] stringr_1.5.1 dplyr_1.1.4 purrr_1.1.0
[7] readr_2.1.5 tidyr_1.3.1 tibble_3.3.0
[10] ggplot2_3.5.2 tidyverse_2.0.0 ANN2_2.3.4
[13] neuralnet_1.44.2 gbm_2.2.2 randomForest_4.7-1.2
[16] rpart_4.1.23 earth_5.3.4 plotmo_3.6.4
[19] plotrix_3.8-4 Formula_1.2-5 sommer_4.4.2
[22] crayon_1.5.3 MASS_7.3-60.2 Matrix_1.7-0
[25] doParallel_1.0.17 iterators_1.0.14 foreach_1.5.2
[28] kableExtra_1.4.0 cvTools_0.3.3 robustbase_0.99-4-1
[31] lattice_0.22-7
loaded via a namespace (and not attached):
[1] pROC_1.18.5 writexl_1.5.4 rlang_1.1.6
[4] magrittr_2.0.3 git2r_0.36.2 compiler_4.4.1
[7] systemfonts_1.2.3 vctrs_0.6.5 reshape2_1.4.4
[10] pkgconfig_2.0.3 fastmap_1.2.0 utf8_1.2.6
[13] promises_1.3.3 rmarkdown_2.29 prodlim_2025.04.28
[16] tzdb_0.5.0 xfun_0.52 cachem_1.1.0
[19] jsonlite_2.0.0 recipes_1.3.1 later_1.4.2
[22] R6_2.6.1 bslib_0.9.0 stringi_1.8.7
[25] RColorBrewer_1.1-3 parallelly_1.45.0 jquerylib_0.1.4
[28] Rcpp_1.1.0 knitr_1.50 future.apply_1.20.0
[31] httpuv_1.6.16 splines_4.4.1 nnet_7.3-19
[34] timechange_0.3.0 tidyselect_1.2.1 rstudioapi_0.17.1
[37] yaml_2.3.10 timeDate_4041.110 codetools_0.2-20
[40] listenv_0.9.1 plyr_1.8.9 withr_3.0.2
[43] evaluate_1.0.4 future_1.58.0 survival_3.6-4
[46] xml2_1.3.8 pillar_1.11.0 whisker_0.4.1
[49] stats4_4.4.1 generics_0.1.4 rprojroot_2.0.4
[52] hms_1.1.3 scales_1.4.0 globals_0.18.0
[55] class_7.3-22 glue_1.8.0 tools_4.4.1
[58] data.table_1.17.8 ModelMetrics_1.2.2.2 gower_1.0.2
[61] fs_1.6.6 grid_4.4.1 ipred_0.9-15
[64] nlme_3.1-168 cli_3.6.5 textshaping_1.0.1
[67] workflowr_1.7.1 viridisLite_0.4.2 svglite_2.2.1
[70] lava_1.8.1 gtable_0.3.6 DEoptimR_1.1-3-1
[73] sass_0.4.10 digest_0.6.37 farver_2.1.2
[76] htmltools_0.5.8.1 lifecycle_1.0.4 hardhat_1.4.1 Weverton Gomes da Costa, Pós-Doutorando, Departamento de Estatística - Universidade Federal de Viçosa, wevertonufv@gmail.com↩︎