Mercurial > repos > ppericard > mixomics_blocksplsda
view Integration_block_splsda_fonc.R @ 4:b0ab97ffc2a1 draft
planemo upload for repository https://github.com/bilille/galaxy-mixomics-blocksplsda commit 0bf5c0745f406f2eca9c708a062c975b1f7ea386
| author | ppericard |
|---|---|
| date | Mon, 09 Mar 2020 11:53:32 -0400 |
| parents | 0a3c83f2197a |
| children |
line wrap: on
line source
# La fonction meanSpotRepl remplace les valeurs des spots répliqués par la # moyenne de leurs intensités. meanSpotRepl <-function(mat) { ProbeName = colnames(mat) isDup = duplicated(ProbeName) dupNames = ProbeName[isDup] for (dups in unique(dupNames)) { mat[dups,] = apply(mat[which(colnames(mat) == dups), ], 2, mean) } res = mat[, -which(isDup)] # On retire de la matrice mat toutes les spots qui sont répliqués. return(res) } # La fonction supprimerVaConst supprime de la matrice mat les variables constantes. supprimerVaConst <-function(mat) { name_mat = deparse(substitute(mat)) cat(paste0("Pour ", name_mat, ", avant suppression des variables constantes, il y a ", dim(mat)[2], " variables."), "\n") indiceVaConst = sapply(1:dim(mat)[2], FUN = function(i){ col_mat_i = mat[, i] res = all(col_mat_i == col_mat_i[1]) return(res) }) if(length(which(indiceVaConst == FALSE)) != 0) { res = mat[, which(indiceVaConst == FALSE)] }else{ res = mat } cat(paste0("Pour ", name_mat, ", après suppression des variables constantes, il reste ", dim(res)[2], " variables."), "\n") return(res) } # La fonction supprimerVaNaInd supprime les variables contenant des NA # (si on garde les NA dans ces variables, les composantes du bloc ne # sont plus orthogonales). supprimerVaNaInd <-function(mat) { indiceVaNaInd = sapply(1:dim(mat)[2], FUN = function(i){ col_mat_i = mat[, i] indNA = length(which(is.na(col_mat_i) == TRUE)) if(indNA >= 1) { res = FALSE }else{ res = TRUE } return(res) }) mat2 = mat[, which(indiceVaNaInd == TRUE)] name_mat = deparse(substitute(mat)) cat(paste0("Pour ", name_mat, ", après suppression des variables contenant des NA, il reste ", dim(mat2)[2], " variables."), "\n") return(mat2) } # La fonction varAnnotation permet de fournir des informations sur les variables # sélectionnées pour un design. varAnnotation <-function(variablesSelect, data_transcripto_col, annot_metageno_caecum) { res_variablesSelect = variablesSelect noms_blocks = sapply(1:length(variablesSelect), FUN = function(i){ ch = strsplit(names(variablesSelect)[i], split = "_")[[1]] if(ch[1] == "resBio") { res = "resBio" }else{ res = paste(ch[1:2], collapse = "_") } return(res) }) ind_transcripto_col = which(noms_blocks == "transcripto_col") if(length(variablesSelect[[ind_transcripto_col]]) != 0) { varSelect_transcripto_colTemp = variablesSelect[[ind_transcripto_col]] varSelect_transcripto_col = sapply(1:length(varSelect_transcripto_colTemp), FUN = function(i){ ch = strsplit(varSelect_transcripto_colTemp[i], split = "_")[[1]] res = paste(ch[2:length(ch)], collapse = "_") return(res) }) dataframe_varSelect_transcripto_col = data.frame(ProbeName = varSelect_transcripto_col) dataframe_annot_transcripto_col = data.frame(ProbeName = data_transcripto_col$genes$ProbeName, GeneName = data_transcripto_col$genes$GeneName, Description = data_transcripto_col$genes$Description, SystematicName = data_transcripto_col$genes$SystematicName) tab_transcripto_col = join(x = dataframe_varSelect_transcripto_col, y = dataframe_annot_transcripto_col, type = "inner", by = "ProbeName") res_variablesSelect[[ind_transcripto_col]] = tab_transcripto_col } ind_metageno_caecum = which(noms_blocks == "metageno_caecum") if(length(variablesSelect[[ind_metageno_caecum]]) != 0) { varSelect_metageno_caecum_Temp1 = variablesSelect[[ind_metageno_caecum]] varSelect_metageno_caecum_Temp2 = sapply(1:length(varSelect_metageno_caecum_Temp1), FUN = function(i){ ch = strsplit(varSelect_metageno_caecum_Temp1[i], split = "")[[1]] if(ch[1] == "X") { res = paste(ch[2:length(ch)], collapse = "") }else{ res = varSelect_metageno_caecum_Temp1[i] } return(res) }) dataframe_varSelect_metageno_caecum = data.frame(taxon = varSelect_metageno_caecum_Temp2) dataframe_annot_metageno_caecum = data.frame(annot_metageno_caecum) colnames(dataframe_annot_metageno_caecum)[1] = "taxon" tab_metageno_caecum = join(x = dataframe_varSelect_metageno_caecum, y = dataframe_annot_metageno_caecum, type = "inner", by = "taxon") res_variablesSelect[[ind_metageno_caecum]] = tab_metageno_caecum } ind_metabo_S1 = which(noms_blocks == "metabo_S1") if(length(variablesSelect[[ind_metabo_S1]]) != 0) { res_variablesSelect[[ind_metabo_S1]] = data.frame(variable = variablesSelect[[ind_metabo_S1]]) } ind_resBio = which(noms_blocks == "resBio") if(length(variablesSelect[[ind_resBio]]) != 0) { res_variablesSelect[[ind_resBio]] = data.frame(variable = variablesSelect[[ind_resBio]]) } return(res_variablesSelect) } # La fonction varAnnotation_gene_6blocks permet de fournir des informations sur les variables # sélectionnées pour un design. varAnnotation_gene_6blocks <-function(variablesSelect, data_transcripto_col, data_transcripto_tae, annot_metageno_caecum, metavar_metaboLC_S1, metavar_resBio, metavar_cyto) { res_variablesSelect = variablesSelect noms_blocks = names(variablesSelect) ind_transcripto_col = which(noms_blocks == "transcripto_col") if(length(ind_transcripto_col) != 0) { varSelect_transcripto_colTemp = variablesSelect[[ind_transcripto_col]] if(length(varSelect_transcripto_colTemp) != 0) { varSelect_transcripto_col = sapply(1:length(varSelect_transcripto_colTemp), FUN = function(i){ variable_i = varSelect_transcripto_colTemp[i] res = gsub("Colon_", "", variable_i, fixed = TRUE) return(res) }) dataframe_varSelect_transcripto_col = data.frame(GeneName = varSelect_transcripto_col) dataframe_annot_transcripto_col = data.frame(ProbeName = data_transcripto_col$genes$ProbeName, GeneName = data_transcripto_col$genes$GeneName, Description = data_transcripto_col$genes$Description, SystematicName = data_transcripto_col$genes$SystematicName) tab_transcripto_col = join(x = dataframe_varSelect_transcripto_col, y = dataframe_annot_transcripto_col, type = "inner", by = "GeneName") res_variablesSelect[[ind_transcripto_col]] = tab_transcripto_col }else{ res_variablesSelect[[ind_transcripto_col]] = "" } } ind_transcripto_tae = which(noms_blocks == "transcripto_tae") if(length(ind_transcripto_tae) != 0) { varSelect_transcripto_taeTemp = variablesSelect[[ind_transcripto_tae]] if(length(varSelect_transcripto_taeTemp) != 0) { varSelect_transcripto_tae = sapply(1:length(varSelect_transcripto_taeTemp), FUN = function(i){ variable_i = varSelect_transcripto_taeTemp[i] res = gsub("TAE_", "", variable_i, fixed = TRUE) return(res) }) dataframe_varSelect_transcripto_tae = data.frame(GeneName = varSelect_transcripto_tae) dataframe_annot_transcripto_tae = data.frame(ProbeName = data_transcripto_tae$genes$ProbeName, GeneName = data_transcripto_tae$genes$GeneName, Description = data_transcripto_tae$genes$Description, SystematicName = data_transcripto_tae$genes$SystematicName) tab_transcripto_tae = join(x = dataframe_varSelect_transcripto_tae, y = dataframe_annot_transcripto_tae, type = "inner", by = "GeneName") res_variablesSelect[[ind_transcripto_tae]] = tab_transcripto_tae }else{ res_variablesSelect[[ind_transcripto_tae]] = "" } } ind_metageno_caecum = which(noms_blocks == "metageno_caecum") if(length(ind_metageno_caecum) != 0) { varSelect_metageno_caecum_Temp1 = variablesSelect[[ind_metageno_caecum]] if(length(varSelect_metageno_caecum_Temp1) != 0) { varSelect_metageno_caecum_Temp2 = sapply(1:length(varSelect_metageno_caecum_Temp1), FUN = function(i){ ch = strsplit(varSelect_metageno_caecum_Temp1[i], split = "")[[1]] if(ch[1] == "X") { res = paste(ch[2:length(ch)], collapse = "") }else{ res = varSelect_metageno_caecum_Temp1[i] } return(res) }) dataframe_varSelect_metageno_caecum = data.frame(taxon = varSelect_metageno_caecum_Temp2) dataframe_annot_metageno_caecum = data.frame(annot_metageno_caecum) colnames(dataframe_annot_metageno_caecum)[1] = "taxon" tab_metageno_caecum = join(x = dataframe_varSelect_metageno_caecum, y = dataframe_annot_metageno_caecum, type = "inner", by = "taxon") res_variablesSelect[[ind_metageno_caecum]] = tab_metageno_caecum }else{ res_variablesSelect[[ind_metageno_caecum]] = "" } } ind_metaboLC_S1 = which(noms_blocks == "metaboLC_S1") if(length(ind_metaboLC_S1) != 0) { if(length(variablesSelect[[ind_metaboLC_S1]]) != 0) { res_variablesSelect[[ind_metaboLC_S1]] = data.frame(variable = variablesSelect[[ind_metaboLC_S1]]) }else{ res_variablesSelect[[ind_metaboLC_S1]] = "" } } ind_resBio = which(noms_blocks == "resBio") if(length(ind_resBio) != 0) { if(length(variablesSelect[[ind_resBio]]) != 0) { res_variablesSelect[[ind_resBio]] = data.frame(variable = variablesSelect[[ind_resBio]]) }else{ res_variablesSelect[[ind_resBio]] = "" } } ind_cyto = which(noms_blocks == "cyto") if(length(ind_cyto) != 0) { if(length(variablesSelect[[ind_cyto]]) != 0) { res_variablesSelect[[ind_cyto]] = data.frame(variable = variablesSelect[[ind_cyto]]) }else{ res_variablesSelect[[ind_cyto]] = "" } } return(res_variablesSelect) } # La fonction varAnnotation_gene_6blocks permet de fournir des informations sur les variables # sélectionnées pour un design. varAnnotation_gene_6blocks <-function(variablesSelect, data_transcripto_col, data_transcripto_tae, annot_metageno_caecum, metavar_metaboLC_S1, metavar_resBio, metavar_cyto) { res_variablesSelect = variablesSelect noms_blocks = names(variablesSelect) ind_transcripto_col = which(noms_blocks == "transcripto_col") if(length(ind_transcripto_col) != 0) { varSelect_transcripto_colTemp = variablesSelect[[ind_transcripto_col]] if(length(varSelect_transcripto_colTemp) != 0) { varSelect_transcripto_col = sapply(1:length(varSelect_transcripto_colTemp), FUN = function(i){ variable_i = varSelect_transcripto_colTemp[i] res = gsub("Colon_", "", variable_i, fixed = TRUE) return(res) }) dataframe_varSelect_transcripto_col = data.frame(GeneName = varSelect_transcripto_col) dataframe_annot_transcripto_col = data.frame(ProbeName = data_transcripto_col$genes$ProbeName, GeneName = data_transcripto_col$genes$GeneName, Description = data_transcripto_col$genes$Description, SystematicName = data_transcripto_col$genes$SystematicName) tab_transcripto_col = join(x = dataframe_varSelect_transcripto_col, y = dataframe_annot_transcripto_col, type = "inner", by = "GeneName") res_variablesSelect[[ind_transcripto_col]] = tab_transcripto_col }else{ res_variablesSelect[[ind_transcripto_col]] = "" } } ind_transcripto_tae = which(noms_blocks == "transcripto_tae") if(length(ind_transcripto_tae) != 0) { varSelect_transcripto_taeTemp = variablesSelect[[ind_transcripto_tae]] if(length(varSelect_transcripto_taeTemp) != 0) { varSelect_transcripto_tae = sapply(1:length(varSelect_transcripto_taeTemp), FUN = function(i){ variable_i = varSelect_transcripto_taeTemp[i] res = gsub("TAE_", "", variable_i, fixed = TRUE) return(res) }) dataframe_varSelect_transcripto_tae = data.frame(GeneName = varSelect_transcripto_tae) dataframe_annot_transcripto_tae = data.frame(ProbeName = data_transcripto_tae$genes$ProbeName, GeneName = data_transcripto_tae$genes$GeneName, Description = data_transcripto_tae$genes$Description, SystematicName = data_transcripto_tae$genes$SystematicName) tab_transcripto_tae = join(x = dataframe_varSelect_transcripto_tae, y = dataframe_annot_transcripto_tae, type = "inner", by = "GeneName") res_variablesSelect[[ind_transcripto_tae]] = tab_transcripto_tae }else{ res_variablesSelect[[ind_transcripto_tae]] = "" } } ind_metageno_caecum = which(noms_blocks == "metageno_caecum") if(length(ind_metageno_caecum) != 0) { varSelect_metageno_caecum_Temp1 = variablesSelect[[ind_metageno_caecum]] if(length(varSelect_metageno_caecum_Temp1) != 0) { varSelect_metageno_caecum_Temp2 = sapply(1:length(varSelect_metageno_caecum_Temp1), FUN = function(i){ ch = strsplit(varSelect_metageno_caecum_Temp1[i], split = "")[[1]] if(ch[1] == "X") { res = paste(ch[2:length(ch)], collapse = "") }else{ res = varSelect_metageno_caecum_Temp1[i] } return(res) }) dataframe_varSelect_metageno_caecum = data.frame(taxon = varSelect_metageno_caecum_Temp2) dataframe_annot_metageno_caecum = data.frame(annot_metageno_caecum) colnames(dataframe_annot_metageno_caecum)[1] = "taxon" tab_metageno_caecum = join(x = dataframe_varSelect_metageno_caecum, y = dataframe_annot_metageno_caecum, type = "inner", by = "taxon") res_variablesSelect[[ind_metageno_caecum]] = tab_metageno_caecum }else{ res_variablesSelect[[ind_metageno_caecum]] = "" } } ind_metaboLC_S1 = which(noms_blocks == "metaboLC_S1") if(length(ind_metaboLC_S1) != 0) { if(length(variablesSelect[[ind_metaboLC_S1]]) != 0) { dataframe_varSelect_metaboLC_S1 = data.frame(Variable = variablesSelect[[ind_metaboLC_S1]]) dataframe_metavar_metaboLC_S1 = data.frame(metavar_metaboLC_S1) dataframe_metavar_varSelect_metaboLC_S1 = join(x = dataframe_varSelect_metaboLC_S1, y = dataframe_metavar_metaboLC_S1, type = "inner", by = "Variable") res_variablesSelect[[ind_metaboLC_S1]] = dataframe_metavar_varSelect_metaboLC_S1 }else{ res_variablesSelect[[ind_metaboLC_S1]] = "" } } ind_resBio = which(noms_blocks == "resBio") if(length(ind_resBio) != 0) { if(length(variablesSelect[[ind_resBio]]) != 0) { dataframe_varSelect_resBio = data.frame(Variable = variablesSelect[[ind_resBio]]) dataframe_metavar_resBio = data.frame(metavar_resBio) dataframe_metavar_varSelect_resBio = join(x = dataframe_varSelect_resBio, y = dataframe_metavar_resBio, type = "inner", by = "Variable") res_variablesSelect[[ind_resBio]] = dataframe_metavar_varSelect_resBio }else{ res_variablesSelect[[ind_resBio]] = "" } } ind_cyto = which(noms_blocks == "cyto") if(length(ind_cyto) != 0) { if(length(variablesSelect[[ind_cyto]]) != 0) { dataframe_varSelect_cyto = data.frame(Variable = variablesSelect[[ind_cyto]]) dataframe_metavar_cyto = data.frame(metavar_cyto) dataframe_metavar_varSelect_cyto = join(x = dataframe_varSelect_cyto, y = dataframe_metavar_cyto, type = "inner", by = "Variable") res_variablesSelect[[ind_cyto]] = dataframe_metavar_varSelect_cyto }else{ res_variablesSelect[[ind_cyto]] = "" } } return(res_variablesSelect) } # La fonction varAnnotation_gene_7blocks permet de fournir des informations sur les variables # sélectionnées pour un design. varAnnotation_gene_7blocks <-function(variablesSelect, data_transcripto_col, data_transcripto_tae, annot_metageno_caecum, metavar_metaboLC_S1, metavar_resBio, metavar_cyto) { res_variablesSelect = variablesSelect noms_blocks = names(variablesSelect) ind_transcripto_col = which(noms_blocks == "transcripto_col") if(length(ind_transcripto_col) != 0) { varSelect_transcripto_colTemp = variablesSelect[[ind_transcripto_col]] if(length(varSelect_transcripto_colTemp) != 0) { varSelect_transcripto_col = sapply(1:length(varSelect_transcripto_colTemp), FUN = function(i){ variable_i = varSelect_transcripto_colTemp[i] res = gsub("Colon_", "", variable_i, fixed = TRUE) return(res) }) dataframe_varSelect_transcripto_col = data.frame(GeneName = varSelect_transcripto_col) dataframe_annot_transcripto_col = data.frame(ProbeName = data_transcripto_col$genes$ProbeName, GeneName = data_transcripto_col$genes$GeneName, Description = data_transcripto_col$genes$Description, SystematicName = data_transcripto_col$genes$SystematicName) tab_transcripto_col = join(x = dataframe_varSelect_transcripto_col, y = dataframe_annot_transcripto_col, type = "inner", by = "GeneName") res_variablesSelect[[ind_transcripto_col]] = tab_transcripto_col }else{ res_variablesSelect[[ind_transcripto_col]] = "" } } ind_transcripto_tae = which(noms_blocks == "transcripto_tae") if(length(ind_transcripto_tae) != 0) { varSelect_transcripto_taeTemp = variablesSelect[[ind_transcripto_tae]] if(length(varSelect_transcripto_taeTemp) != 0) { varSelect_transcripto_tae = sapply(1:length(varSelect_transcripto_taeTemp), FUN = function(i){ variable_i = varSelect_transcripto_taeTemp[i] res = gsub("TAE_", "", variable_i, fixed = TRUE) return(res) }) dataframe_varSelect_transcripto_tae = data.frame(GeneName = varSelect_transcripto_tae) dataframe_annot_transcripto_tae = data.frame(ProbeName = data_transcripto_tae$genes$ProbeName, GeneName = data_transcripto_tae$genes$GeneName, Description = data_transcripto_tae$genes$Description, SystematicName = data_transcripto_tae$genes$SystematicName) tab_transcripto_tae = join(x = dataframe_varSelect_transcripto_tae, y = dataframe_annot_transcripto_tae, type = "inner", by = "GeneName") res_variablesSelect[[ind_transcripto_tae]] = tab_transcripto_tae }else{ res_variablesSelect[[ind_transcripto_tae]] = "" } } ind_metageno_caecum = which(noms_blocks == "metageno_caecum") if(length(ind_metageno_caecum) != 0) { varSelect_metageno_caecum_Temp1 = variablesSelect[[ind_metageno_caecum]] if(length(varSelect_metageno_caecum_Temp1) != 0) { varSelect_metageno_caecum_Temp2 = sapply(1:length(varSelect_metageno_caecum_Temp1), FUN = function(i){ ch = strsplit(varSelect_metageno_caecum_Temp1[i], split = "")[[1]] if(ch[1] == "X") { res = paste(ch[2:length(ch)], collapse = "") }else{ res = varSelect_metageno_caecum_Temp1[i] } return(res) }) dataframe_varSelect_metageno_caecum = data.frame(taxon = varSelect_metageno_caecum_Temp2) dataframe_annot_metageno_caecum = data.frame(annot_metageno_caecum) colnames(dataframe_annot_metageno_caecum)[1] = "taxon" tab_metageno_caecum = join(x = dataframe_varSelect_metageno_caecum, y = dataframe_annot_metageno_caecum, type = "inner", by = "taxon") res_variablesSelect[[ind_metageno_caecum]] = tab_metageno_caecum }else{ res_variablesSelect[[ind_metageno_caecum]] = "" } } ind_metaboLC_S1 = which(noms_blocks == "metaboLC_S1") if(length(ind_metaboLC_S1) != 0) { if(length(variablesSelect[[ind_metaboLC_S1]]) != 0) { dataframe_varSelect_metaboLC_S1 = data.frame(Variable = variablesSelect[[ind_metaboLC_S1]]) dataframe_metavar_metaboLC_S1 = data.frame(metavar_metaboLC_S1) dataframe_metavar_varSelect_metaboLC_S1 = join(x = dataframe_varSelect_metaboLC_S1, y = dataframe_metavar_metaboLC_S1, type = "inner", by = "Variable") res_variablesSelect[[ind_metaboLC_S1]] = dataframe_metavar_varSelect_metaboLC_S1 }else{ res_variablesSelect[[ind_metaboLC_S1]] = "" } } ind_metaboGC_S1 = which(noms_blocks == "metaboGC_S1") if(length(ind_metaboGC_S1) != 0) { if(length(variablesSelect[[ind_metaboGC_S1]]) != 0) { res_variablesSelect[[ind_metaboGC_S1]] = data.frame(Variable = variablesSelect[[ind_metaboGC_S1]]) }else{ res_variablesSelect[[ind_metaboGC_S1]] = "" } } ind_resBio = which(noms_blocks == "resBio") if(length(ind_resBio) != 0) { if(length(variablesSelect[[ind_resBio]]) != 0) { dataframe_varSelect_resBio = data.frame(Variable = variablesSelect[[ind_resBio]]) dataframe_metavar_resBio = data.frame(metavar_resBio) dataframe_metavar_varSelect_resBio = join(x = dataframe_varSelect_resBio, y = dataframe_metavar_resBio, type = "inner", by = "Variable") res_variablesSelect[[ind_resBio]] = dataframe_metavar_varSelect_resBio }else{ res_variablesSelect[[ind_resBio]] = "" } } ind_cyto = which(noms_blocks == "cyto") if(length(ind_cyto) != 0) { if(length(variablesSelect[[ind_cyto]]) != 0) { dataframe_varSelect_cyto = data.frame(Variable = variablesSelect[[ind_cyto]]) dataframe_metavar_cyto = data.frame(metavar_cyto) dataframe_metavar_varSelect_cyto = join(x = dataframe_varSelect_cyto, y = dataframe_metavar_cyto, type = "inner", by = "Variable") res_variablesSelect[[ind_cyto]] = dataframe_metavar_varSelect_cyto }else{ res_variablesSelect[[ind_cyto]] = "" } } return(res_variablesSelect) } # Integration 6 blocs norm sonde ------------------------------------------ # La fonction plotVarZoom permet de zoomer sur le cercle de corrélation et de récupérer les variables # contenues dans ce rectangle. plotVarZoom <-function(res_block_splsda, comp = 1:2, blocks, block_Y = NULL, vec_col, cutoff, rad.in = 0.5, min.X = -1, max.X = 1, min.Y = -1, max.Y = 1, cex = 0.7, cex_legend = 0.8, pos = c(1.2, 0), pch = 20, inset = c(-0.25, 0)) { if(class(res_block_splsda)[1] == "block.splsda") { circle = list() circle[[1]] = ellipse(0, levels = 1, t = 1) circle[[2]] = ellipse(0, levels = 1, t = rad.in) circle = data.frame(do.call("rbind", circle), "Circle" = c(rep("Main circle", 100), rep("Inner circle", 100))) MainCircle = circle[grep("Main circle", circle[, 3]), ] InnerCircle = circle[grep("Inner circle", circle[, 3]), ] if(length(blocks) > 1) { noms_bloc = names(res_block_splsda$variates) mat_comp1 = sapply(blocks, FUN = function(i){ res = res_block_splsda$variates[[i]][, 1] return(res) }) colnames(mat_comp1) = noms_bloc[blocks] mat_cor_comp1 = cor(mat_comp1) mat_comp2 = sapply(blocks, FUN = function(i){ res = res_block_splsda$variates[[i]][, 2] return(res) }) colnames(mat_comp2) = noms_bloc[blocks] mat_cor_comp2 = cor(mat_comp2) } # Fin if(length(blocks) > 1). # Pour chaque bloc, calcul des corrélations entre la première # composante et les variables sélectionnées et les corrélations entre # la deuxième composante et les variables sélectionnées. Pour la réponse, # calcul de la corrélation entre les variables de la réponse et la première composante # du premier bloc sélectionné et de la corrélation entre les variables de # la réponse et la deuxième composante du premier bloc sélectionné. liste_matCor_comp_var = list() varSelect_comp1 = selectVar(res_block_splsda, comp = comp[1]) varSelect_comp2 = selectVar(res_block_splsda, comp = comp[2]) vec_nom_blockEtReponse = c() blocksEtReponse = c(blocks, which(res_block_splsda$names$blocks == "Y")) for(i in 1:length(blocksEtReponse)) { indice_block_i = blocksEtReponse[i] nom_blockEtReponse_i = res_block_splsda$names$blocks[indice_block_i] if(nom_blockEtReponse_i == "Y") { if(!is.null(block_Y)) { block_i = block_Y comp1 = res_block_splsda$variates[[blocks[1]]][, comp[1]] comp2 = res_block_splsda$variates[[blocks[1]]][, comp[2]] vecCor_comp1_var = sapply(1:dim(block_i)[2], FUN = function(j){ cor(comp1, block_i[, j], use = "complete.obs") }) vecCor_comp2_var = sapply(1:dim(block_i)[2], FUN = function(j){ cor(comp2, block_i[, j], use = "complete.obs") }) }else{ cat("La réponse n'est pas saisie comme paramètre d'entrée", "\n") } }else{ comp1 = res_block_splsda$variates[[indice_block_i]][, comp[1]] comp2 = res_block_splsda$variates[[indice_block_i]][, comp[2]] varSelect_comp1_i = varSelect_comp1[[indice_block_i]][[1]] varSelect_comp2_i = varSelect_comp2[[indice_block_i]][[1]] varSelect_i = unique(c(varSelect_comp1_i, varSelect_comp2_i)) block_i = res_block_splsda$X[[indice_block_i]][, varSelect_i] if(i == 1) { vecCor_comp1_var = sapply(1:dim(block_i)[2], FUN = function(j){ cor(comp1, block_i[, j], use = "complete.obs") }) vecCor_comp2_var = sapply(1:dim(block_i)[2], FUN = function(j){ cor(comp2, block_i[, j], use = "complete.obs") }) }else{ signeCor_comp1 = sign(mat_cor_comp1[1, indice_block_i]) vecCor_comp1_var = sapply(1:dim(block_i)[2], FUN = function(j){ res = signeCor_comp1*cor(comp1, block_i[, j], use = "complete.obs") return(res) }) signeCor_comp2 = sign(mat_cor_comp2[1, indice_block_i]) vecCor_comp2_var = sapply(1:dim(block_i)[2], FUN = function(j){ res = signeCor_comp2*cor(comp2, block_i[, j], use = "complete.obs") return(res) }) } } matCor_comp_var = rbind(vecCor_comp1_var, vecCor_comp2_var) colnames(matCor_comp_var) = colnames(block_i) liste_matCor_comp_var[[i]] = matCor_comp_var vec_nom_blockEtReponse_i = rep(nom_blockEtReponse_i, dim(block_i)[2]) vec_nom_blockEtReponse = c(vec_nom_blockEtReponse, vec_nom_blockEtReponse_i) } # Fin for(i in 1:length(blocks)). matCor_Allcomp_Allvar = t(Reduce(cbind, liste_matCor_comp_var)) # indice permet de récupérer les variables de chaque bloc fortement corrélées avec soit # la première composante ou la deuxième composante dans une partie du cercle de corrélation # et de récupérer la variable réponse dans une partie du cercle de corrélation. indice = sapply(1:dim(matCor_Allcomp_Allvar)[1], FUN = function(k){ cor1 = matCor_Allcomp_Allvar[k, 1] cor2 = matCor_Allcomp_Allvar[k, 2] blockEtReponse_k = vec_nom_blockEtReponse[k] if(blockEtReponse_k == "Y") { cond2 = cor1 > min.X & cor1 < max.X & cor2 > min.Y & cor2 < max.Y }else{ cond1 = abs(cor1) > cutoff | abs(cor2) > cutoff cond2 = cor1 > min.X & cor1 < max.X & cor2 > min.Y & cor2 < max.Y & cond1 } return(cond2) }) matCor_Allcomp_AllvarSelect = matCor_Allcomp_Allvar[indice, , drop = FALSE] varSelect = rownames(matCor_Allcomp_AllvarSelect) dataframe_Cor_Allcomp_Allvar = data.frame(cbind(rownames(matCor_Allcomp_Allvar), vec_nom_blockEtReponse, matCor_Allcomp_Allvar)) colnames(dataframe_Cor_Allcomp_Allvar) = c("variable", "bloc", "cor_comp1_var", "cor_comp2_var") dataframe_Cor_Allcomp_Allvar[, 1:2] = apply(dataframe_Cor_Allcomp_Allvar[, 1:2], 2, as.character) dataframe_Cor_Allcomp_AllvarSelect = dataframe_Cor_Allcomp_Allvar[indice, ] # Tracé de la superposition des cerles de corrélation. plot(MainCircle[, 1], MainCircle[, 2], type = "l", xlab = paste0("composante ", comp[1]), ylab = paste0("composante ", comp[2])) points(InnerCircle[, 1], InnerCircle[, 2], type = "l") if(dim(matCor_Allcomp_AllvarSelect)[1] != 0) { nom_blockEtReponseSelect = unique(dataframe_Cor_Allcomp_AllvarSelect[, 2]) indice_blockEtReponseSelect = sapply(1:length(nom_blockEtReponseSelect), FUN = function(i){ ind = which(res_block_splsda$names$blocks == nom_blockEtReponseSelect[i]) if(length(ind) != 0) { res = ind }else{ res = which(res_block_splsda$names$blocks == "Y") } return(res) }) vec_colSelect = vec_col[indice_blockEtReponseSelect] if(length(blocksEtReponse) == 1) { points(matCor_Allcomp_AllvarSelect[, 1], matCor_Allcomp_AllvarSelect[, 2], col = NULL) text(matCor_Allcomp_AllvarSelect[, 1], matCor_Allcomp_AllvarSelect[, 2], labels = rownames(matCor_Allcomp_AllvarSelect), cex = cex, col = vec_colSelect[1]) }else{ nbVarSelect_bloc = cumsum(sapply(1:length(nom_blockEtReponseSelect), FUN = function(j){ res = length(which(dataframe_Cor_Allcomp_AllvarSelect[, 2] == nom_blockEtReponseSelect[j])) return(res) })) for(i in 1:length(nbVarSelect_bloc)) { if(i == 1) { indice1 = 1:nbVarSelect_bloc[1] if(length(indice1) != 0) { matCor_Allcomp_AllvarSelect2 = matCor_Allcomp_AllvarSelect[indice1, , drop = FALSE] points(matCor_Allcomp_AllvarSelect2[, 1], matCor_Allcomp_AllvarSelect2[, 2], col = NULL) text(matCor_Allcomp_AllvarSelect2[, 1], matCor_Allcomp_AllvarSelect2[, 2], labels = rownames(matCor_Allcomp_AllvarSelect2), cex = cex, col = rep(vec_colSelect[i], dim(matCor_Allcomp_AllvarSelect2)[1])) }else{ cat(paste0("Il n'y a de variables dans cette zone du cercle de corrélation pour le bloc ", nom_blockEtReponseSelect[i]), "\n") } }else{ indice2 = (nbVarSelect_bloc[i - 1] + 1):nbVarSelect_bloc[i] if(length(indice2) != 0) { matCor_Allcomp_AllvarSelect2 = matCor_Allcomp_AllvarSelect[indice2, , drop = FALSE] points(matCor_Allcomp_AllvarSelect2[, 1], matCor_Allcomp_AllvarSelect2[, 2], col = NULL) text(matCor_Allcomp_AllvarSelect2[, 1], matCor_Allcomp_AllvarSelect2[, 2], labels = rownames(matCor_Allcomp_AllvarSelect2), cex = cex, col = rep(vec_colSelect[i], dim(matCor_Allcomp_AllvarSelect2)[1])) }else{ cat(paste0("Il n'y a de variables dans cette zone du cercle de corrélation pour le bloc ", nom_blockEtReponseSelect[i]), "\n") } } } } par(xpd = TRUE) legend(x = pos[1], y = pos[2], legend = nom_blockEtReponseSelect, pch = pch, col = vec_colSelect, cex = cex_legend, inset = inset) }else{ cat("Il n'y a de variables dans cette zone du cercle de corrélation", "\n") } # Suppression de dataframe_Cor_Allcomp_Allvar, dataframe_Cor_Allcomp_AllvarSelect et varSelect de la variable # réponse. if(!is.null(block_Y)) { for(i in 1:dim(block_Y)[2]) { variableReponse = colnames(block_Y)[i] # dataframe_Cor_Allcomp_Allvar if(dim(dataframe_Cor_Allcomp_Allvar)[1] != 0) { ind_Allvar = which(dataframe_Cor_Allcomp_Allvar$variable == variableReponse) if(length(ind_Allvar) != 0) { dataframe_Cor_Allcomp_Allvar = dataframe_Cor_Allcomp_Allvar[- ind_Allvar, ] } } # dataframe_Cor_Allcomp_AllvarSelect if(dim(dataframe_Cor_Allcomp_AllvarSelect)[1] != 0) { ind_AllvarSelect = which(dataframe_Cor_Allcomp_AllvarSelect$variable == variableReponse) if(length(ind_AllvarSelect) != 0) { dataframe_Cor_Allcomp_AllvarSelect = dataframe_Cor_Allcomp_AllvarSelect[- ind_AllvarSelect, ] } } # varSelect if(length(varSelect) != 0) { ind_varSelect = which(varSelect == variableReponse) if(length(ind_varSelect) != 0) { varSelect = varSelect[- ind_varSelect] } } } # Fin for(i in 1:dim(block_Y)[2]). } res = list(dataframe_Cor_Allcomp_Allvar = dataframe_Cor_Allcomp_Allvar, dataframe_Cor_Allcomp_AllvarSelect = dataframe_Cor_Allcomp_AllvarSelect, varSelect = varSelect) return(res) }else{ cat("Erreur : il ne s'agit pas de la sortie de la fonction block.splsda", "\n") } } # La fonction networkVarSelect permet de tracer un réseau pour les variables de # certains blocs. networkVarSelect <-function(object, mat_Y, comp = 1:2, listeVar, blocks, cutoff = 0 ) { if(class(object)[1] == "block.splsda") { nomBlocs = names(listeVar) liste_XSelect = lapply(1:length(listeVar), FUN = function(i){ nomBloc_i = names(listeVar)[i] ind1 = which(object$names$blocks == nomBloc_i) ind2 = colnames(object$X[[ind1]])%in%listeVar[[i]] res = object$X[[ind1]][, ind2, drop = FALSE] return(res) }) names(liste_XSelect) = nomBlocs liste_matSelect = liste_XSelect if(!is.null(mat_Y)) { liste_matSelect = c(liste_matSelect, list(mat_Y)) names(liste_matSelect)[length(liste_matSelect)] = "Y" } # compute the similarity between var1 of block1 and var2 of block2. coord = list() for(k in 1:length(blocks)) { nomBloc_k = names(liste_matSelect)[k] mat_k = liste_matSelect[[k]][, ,drop = FALSE] if(nomBloc_k == "Y") { coord[[k]] = cor(mat_k, object$variates[[blocks[1]]][, comp]) }else{ coord[[k]] = cor(mat_k, object$variates[[blocks[k]]][, comp]) } if(dim(mat_k)[2] == 1) { coord[[k]] = as.matrix(coord[[k]], nrow = 1) rownames(coord[[k]]) = colnames(mat_k) } } # Fin for(k in 1:length(blocks)). l = 1 M_block = list() node.X1 = node.X2 = w = NULL for(j in 1:(length(blocks) - 1)) { for(k in (j + 1):length(blocks)) { M_block[[l]] = coord[[j]][, comp, drop = FALSE] %*% t(coord[[k]][, comp, drop = FALSE]) X1 = rownames(coord[[j]]) X2 = rownames(coord[[k]]) rep.X1 = rep(X1, each = length(X2)) rep.X2 = rep(X2, length(X1)) node.X1= c(node.X1, rep.X1) node.X2 = c(node.X2, rep.X2) w = c(w, as.vector(t(M_block[[l]]))) l = l + 1 } # Fin for(k in (j + 1):length(blocks)). } # Fin for(j in 1:(length(blocks) - 1)). # nodes group = NULL temp = lapply(liste_matSelect, function(x) colnames(x)) for (i in 1:length(temp)) { group = c(group, rep(names(temp)[i], length(temp[[i]]))) } # Fin for (i in 1:length(temp)). # nodes nodes = data.frame(name = unlist(temp), group = group) # gR relations = data.frame(from = node.X1, to = node.X2, weight = w) idx = (abs(w) >= cutoff) relations = relations[idx, , drop = FALSE] gR = graph.data.frame(relations, directed = FALSE, vertices = nodes) block.var.names = sapply(1:length(liste_matSelect), FUN = function(i){ res = colnames(liste_matSelect[[i]]) return(res) }) V(gR)$label = unlist(block.var.names) gR = delete.vertices(gR, which(degree(gR) == 0)) res = list(gR = gR) l = 1 for (i in 1:(length(blocks)-1)) { for (j in (i + 1):length(blocks)) { res[paste("M", names(liste_matSelect)[i], names(liste_matSelect)[j], sep="_")] = list(M_block[[l]]) l = l + 1 } # Fin for (j in (i + 1):length(blocks)). } # Fin for (i in 1:(length(blocks)-1). res$cutoff = cutoff return(res) } # Fin if(class(object)[1] == "block.splsda"). } # La fonction varAnnotation_6blocks permet de fournir des informations sur les variables # sélectionnées pour un design. varAnnotation_6blocks <-function(variablesSelect, data_transcripto_col, data_transcripto_tae, annot_metageno_caecum) { res_variablesSelect = variablesSelect noms_blocks = names(variablesSelect) ind_transcripto_col = which(noms_blocks == "transcripto_col") if(length(variablesSelect[[ind_transcripto_col]]) != 0) { varSelect_transcripto_colTemp = variablesSelect[[ind_transcripto_col]] varSelect_transcripto_col = sapply(1:length(varSelect_transcripto_colTemp), FUN = function(i){ ch = strsplit(varSelect_transcripto_colTemp[i], split = "_")[[1]] res = paste(ch[2:length(ch)], collapse = "_") return(res) }) dataframe_varSelect_transcripto_col = data.frame(ProbeName = varSelect_transcripto_col) dataframe_annot_transcripto_col = data.frame(ProbeName = data_transcripto_col$genes$ProbeName, GeneName = data_transcripto_col$genes$GeneName, Description = data_transcripto_col$genes$Description, SystematicName = data_transcripto_col$genes$SystematicName) tab_transcripto_col = join(x = dataframe_varSelect_transcripto_col, y = dataframe_annot_transcripto_col, type = "inner", by = "ProbeName") res_variablesSelect[[ind_transcripto_col]] = tab_transcripto_col } ind_transcripto_tae = which(noms_blocks == "transcripto_tae") if(length(variablesSelect[[ind_transcripto_tae]]) != 0) { varSelect_transcripto_taeTemp = variablesSelect[[ind_transcripto_tae]] varSelect_transcripto_tae = sapply(1:length(varSelect_transcripto_taeTemp), FUN = function(i){ ch = strsplit(varSelect_transcripto_taeTemp[i], split = "_")[[1]] res = paste(ch[2:length(ch)], collapse = "_") return(res) }) dataframe_varSelect_transcripto_tae = data.frame(ProbeName = varSelect_transcripto_tae) dataframe_annot_transcripto_tae = data.frame(ProbeName = data_transcripto_tae$genes$ProbeName, GeneName = data_transcripto_tae$genes$GeneName, Description = data_transcripto_tae$genes$Description, SystematicName = data_transcripto_tae$genes$SystematicName) tab_transcripto_tae = join(x = dataframe_varSelect_transcripto_tae, y = dataframe_annot_transcripto_tae, type = "inner", by = "ProbeName") res_variablesSelect[[ind_transcripto_tae]] = tab_transcripto_tae } ind_metageno_caecum = which(noms_blocks == "metageno_caecum") if(length(variablesSelect[[ind_metageno_caecum]]) != 0) { varSelect_metageno_caecum_Temp1 = variablesSelect[[ind_metageno_caecum]] varSelect_metageno_caecum_Temp2 = sapply(1:length(varSelect_metageno_caecum_Temp1), FUN = function(i){ ch = strsplit(varSelect_metageno_caecum_Temp1[i], split = "")[[1]] if(ch[1] == "X") { res = paste(ch[2:length(ch)], collapse = "") }else{ res = varSelect_metageno_caecum_Temp1[i] } return(res) }) dataframe_varSelect_metageno_caecum = data.frame(taxon = varSelect_metageno_caecum_Temp2) dataframe_annot_metageno_caecum = data.frame(annot_metageno_caecum) colnames(dataframe_annot_metageno_caecum)[1] = "taxon" tab_metageno_caecum = join(x = dataframe_varSelect_metageno_caecum, y = dataframe_annot_metageno_caecum, type = "inner", by = "taxon") res_variablesSelect[[ind_metageno_caecum]] = tab_metageno_caecum } ind_metaboLC_S1 = which(noms_blocks == "metaboLC_S1") if(length(variablesSelect[[ind_metaboLC_S1]]) != 0) { res_variablesSelect[[ind_metaboLC_S1]] = data.frame(variable = variablesSelect[[ind_metaboLC_S1]]) } ind_resBio = which(noms_blocks == "resBio") if(length(variablesSelect[[ind_resBio]]) != 0) { res_variablesSelect[[ind_resBio]] = data.frame(variable = variablesSelect[[ind_resBio]]) } ind_cyto = which(noms_blocks == "cyto") if(length(variablesSelect[[ind_cyto]]) != 0) { res_variablesSelect[[ind_cyto]] = data.frame(variable = variablesSelect[[ind_cyto]]) } return(res_variablesSelect) } # Integration 6 blocs norm gene ------------------------------------------- # La fonction matriceCorrelation_comp calcule la matrice de corrélation entre les comp[1] # composantes de chaque bloc et la matrice de corrélation entre les comp[2] composantes de # chaque bloc. matriceCorrelation_comp <-function(res_block_splsda, comp = 1:2) { vec_blocksTemp = res_block_splsda$names$blocks ind_Y = which(vec_blocksTemp == "Y") vec_blocks = vec_blocksTemp[- ind_Y] vec_indice_blocks = sapply(1:length(vec_blocks), FUN = function(i){ res = which(res_block_splsda$names$blocks == vec_blocks[i]) return(res) }) # Calcul de la matrice de corrélations entre les comp[1] composantes de chaque bloc # et de la matrice de corrélation entre les comp[2] composantes de chaque bloc. mat_comp1 = sapply(1:length(vec_indice_blocks), FUN = function(i){ res = res_block_splsda$variates[[vec_indice_blocks[i]]][, comp[1]] return(res) }) colnames(mat_comp1) = vec_blocks mat_cor_comp1 = cor(mat_comp1) mat_comp2 = sapply(1:length(vec_indice_blocks), FUN = function(i){ res = res_block_splsda$variates[[vec_indice_blocks[i]]][, comp[2]] return(res) }) colnames(mat_comp2) = vec_blocks mat_cor_comp2 = cor(mat_comp2) return(list(mat_cor_comp1 = mat_cor_comp1, mat_cor_comp2 = mat_cor_comp2)) } # La fonction matCor détermine toutes les combinaisons possibles des blocs # dont nous pouvons superposer les cercles de corrélation. Cette fonction calcule, # pour chaque variable (variable d'un bloc ou une variable réponse), la corrélation de # cette variable avec la première composante et la corrélation de cette variable avec la # deuxième composante. matCor <-function(res_block_splsda, mat_cor_comp1, mat_cor_comp2, block_Y, comp = 1:2, cutoff_comp) { noms_blocTemp = res_block_splsda$names$blocks ind_Y = which(noms_blocTemp == "Y") noms_bloc = noms_blocTemp[- ind_Y] # Détermination de toutes les combinaisons possibles de blocs dont nous pouvons superposer # les cercles de corrélation. blockSelect = unique(lapply(1:dim(mat_cor_comp1)[1], FUN = function(i){ col_mat_cor_comp1_i = mat_cor_comp1[, i] col_mat_cor_comp2_i = mat_cor_comp2[, i] resultat = c() for(j in 1:length(col_mat_cor_comp1_i)) { cond = abs(col_mat_cor_comp1_i[j]) > cutoff_comp & abs(col_mat_cor_comp2_i[j]) > cutoff_comp if(cond) { resultat = c(resultat, j) } } return(resultat) })) liste_noms_blocks = list() for(i in 1:length(blockSelect)) { blockSelect_i = blockSelect[[i]] for(k in 1:length(blockSelect_i)) { matComb = combn(blockSelect_i, m = k) res = lapply(1:dim(matComb)[2], FUN = function(i){ comb_i = matComb[, i] resultat = noms_bloc[comb_i] return(resultat) }) for(j in 1:length(res)) { liste_noms_blocks = c(liste_noms_blocks, list(res[[j]])) } # Fin for(j in 1:length(liste_noms_blocks)). } # Fin for(k in 1:length(blockSelect_i)). } # Fin for(i in 1:length(blockSelect)). liste_noms_blocks = unique(liste_noms_blocks) # Pour chaque bloc, calcul des corrélations entre la première # composante et les variables sélectionnées et les corrélations entre # la deuxième composante et les variables sélectionnées. Pour la réponse, # calcul de la corrélation entre les variables de la réponse et la première composante # du premier bloc sélectionné et de la corrélation entre les variables de # la réponse et la deuxième composante du premier bloc sélectionné. liste_matCor_comp_var_all = list() varSelect_comp1 = selectVar(res_block_splsda, comp = comp[1]) varSelect_comp2 = selectVar(res_block_splsda, comp = comp[2]) blockSelectEtReponse = list() for(i in 1:length(blockSelect)) { blockSelectEtReponse[[i]] = c(blockSelect[[i]], which(res_block_splsda$names$blocks == "Y")) } # Fin for(i in 1:length(blockSelect)). for(i in 1:length(blockSelectEtReponse)) { blockSelectEtReponse_i = blockSelectEtReponse[[i]] vec_nom_blockEtReponse = c() liste_matCor_comp_var = list() for(j in 1:length(blockSelectEtReponse_i)) { indice_block_i_j = blockSelectEtReponse_i[j] nom_blockEtReponse_i_j = res_block_splsda$names$blocks[indice_block_i_j] if(nom_blockEtReponse_i_j == "Y") { if(!is.null(block_Y)) { block_i_j = block_Y comp1 = res_block_splsda$variates[[blockSelectEtReponse_i[1]]][, comp[1]] comp2 = res_block_splsda$variates[[blockSelectEtReponse_i[1]]][, comp[2]] vecCor_comp1_var = sapply(1:dim(block_Y)[2], FUN = function(j){ cor(comp1, block_Y[, j], use = "complete.obs") }) vecCor_comp2_var = sapply(1:dim(block_Y)[2], FUN = function(j){ cor(comp2, block_Y[, j], use = "complete.obs") }) }else{ cat("La réponse n'est pas saisie comme paramètre d'entrée", "\n") } }else{ comp1 = res_block_splsda$variates[[indice_block_i_j]][, comp[1]] comp2 = res_block_splsda$variates[[indice_block_i_j]][, comp[2]] varSelect_comp1_i_j = varSelect_comp1[[indice_block_i_j]][[1]] varSelect_comp2_i_j = varSelect_comp2[[indice_block_i_j]][[1]] varSelect_i_j = unique(c(varSelect_comp1_i_j, varSelect_comp2_i_j)) block_i_j = res_block_splsda$X[[indice_block_i_j]][, varSelect_i_j] if(i == 1) { vecCor_comp1_var = sapply(1:dim(block_i_j)[2], FUN = function(j){ cor(comp1, block_i_j[, j], use = "complete.obs") }) vecCor_comp2_var = sapply(1:dim(block_i_j)[2], FUN = function(j){ cor(comp2, block_i_j[, j], use = "complete.obs") }) }else{ signeCor_comp1 = sign(mat_cor_comp1[1, indice_block_i_j]) vecCor_comp1_var = sapply(1:dim(block_i_j)[2], FUN = function(j){ res = signeCor_comp1*cor(comp1, block_i_j[, j], use = "complete.obs") return(res) }) signeCor_comp2 = sign(mat_cor_comp2[1, indice_block_i_j]) vecCor_comp2_var = sapply(1:dim(block_i_j)[2], FUN = function(j){ res = signeCor_comp2*cor(comp2, block_i_j[, j], use = "complete.obs") return(res) }) } } matCor_comp_var = rbind(vecCor_comp1_var, vecCor_comp2_var) colnames(matCor_comp_var) = colnames(block_i_j) liste_matCor_comp_var[[j]] = matCor_comp_var vec_nom_blockEtReponse_i_j = rep(nom_blockEtReponse_i_j, dim(block_i_j)[2]) vec_nom_blockEtReponse = c(vec_nom_blockEtReponse, vec_nom_blockEtReponse_i_j) } # Fin for(j in 1:length(blockSelectEtReponse_i)). matCor_Allcomp_Allvar = t(Reduce(cbind, liste_matCor_comp_var)) dataframe_Cor_Allcomp_Allvar = data.frame(cbind(rownames(matCor_Allcomp_Allvar), vec_nom_blockEtReponse, matCor_Allcomp_Allvar)) colnames(dataframe_Cor_Allcomp_Allvar) = c("variable", "bloc", "cor_comp1_var", "cor_comp2_var") liste_matCor_comp_var_all[[i]] = dataframe_Cor_Allcomp_Allvar } # Fin for(i in 1:length(blocks)). names(liste_matCor_comp_var_all) = sapply(1:length(blockSelectEtReponse), FUN = function(i){ blockSelectEtReponse_i = blockSelectEtReponse[[i]] blocks_i = blockSelectEtReponse_i[blockSelectEtReponse_i != ind_Y] nomsBlocks = paste(res_block_splsda$names$blocks[blocks_i], collapse = "_") return(nomsBlocks) }) return(list(liste_matCor_comp_var_all = liste_matCor_comp_var_all, liste_noms_blocks = liste_noms_blocks)) } # La fonction circleCorZoom permet superposer des cercles de corrélations et de zoomer un rectangle du cercle de corrélations. # Cette fonction permet de récupérer les variables contenues dans ce rectangle. circleCorZoom <-function(dataframe_Cor_Allcomp_Allvar, mat_cor_comp1, mat_cor_comp2, nomsBlock, comp = 1:2, cutoff = 0.85, cutoff_comp = 0.8, min.X = -1, max.X = 1, min.Y = -1, max.Y = 1, vec_col = colorRampPalette(brewer.pal(9, "Spectral"))(length(unique(dataframe_Cor_Allcomp_Allvar$bloc))), rad.in = 0.5, cex = 0.7, cex_legend = 0.8, pos = c(1.2, 0), pch = 20, inset = c(-0.25, 0)) { # Nous vérifions que nous pouvons superposer les cercles de corrélation. blockSelect = unique(lapply(1:dim(mat_cor_comp1)[1], FUN = function(i){ col_mat_cor_comp1_i = mat_cor_comp1[, i] col_mat_cor_comp2_i = mat_cor_comp2[, i] resultat = c() for(j in 1:length(col_mat_cor_comp1_i)) { cond = abs(col_mat_cor_comp1_i[j]) > cutoff_comp & abs(col_mat_cor_comp2_i[j]) > cutoff_comp if(cond) { resultat = c(resultat, j) } } return(resultat) })) indice_block_nomsBlock = sapply(1:length(nomsBlock), FUN = function(i){ res = which(colnames(mat_cor_comp1) == nomsBlock[i]) return(res) }) boolean = FALSE i = 1 while(i <=length(blockSelect) & !boolean) { blockSelect_i = blockSelect[[i]] cond = length(which(blockSelect_i%in%indice_block_nomsBlock == TRUE)) == length(nomsBlock) if(cond) { boolean = TRUE } i = i + 1 } # Fin while(i <=length(blockSelect) & !boolean). varSelect = NULL if(!boolean) { cat(paste0("Erreur : les blocs : ", paste(nomsBlock, collapse = ", "), " ne peuvent pas être superposés.", "\n")) }else{ nomsBlockEtReponse = c(nomsBlock, "Y") indice1 = dataframe_Cor_Allcomp_Allvar$bloc%in%nomsBlockEtReponse dataframe_Cor_Allcomp_Allvar2 = dataframe_Cor_Allcomp_Allvar[indice1, ] matCor_Allcomp_Allvar = apply(dataframe_Cor_Allcomp_Allvar2[, 3:4], 2, as.numeric) rownames(matCor_Allcomp_Allvar) = dataframe_Cor_Allcomp_Allvar2$variable # indice permet de récupérer les variables de chaque bloc fortement corrélées avec soit # la première composante ou la deuxième composante dans une partie du cercle de corrélation # et de récupérer la variable réponse. indice = sapply(1:dim(dataframe_Cor_Allcomp_Allvar2)[1], FUN = function(k){ cor1 = matCor_Allcomp_Allvar[k, 1] cor2 = matCor_Allcomp_Allvar[k, 2] blockEtReponse_k = dataframe_Cor_Allcomp_Allvar2[k, 2] if(blockEtReponse_k == "Y") { cond2 = TRUE }else{ cond1 = abs(cor1) > cutoff | abs(cor2) > cutoff cond2 = cor1 > min.X & cor1 < max.X & cor2 > min.Y & cor2 < max.Y & cond1 } return(cond2) }) dataframe_Cor_Allcomp_Allvar2[, 1:2] = apply(dataframe_Cor_Allcomp_Allvar2[, 1:2], 2, as.character) dataframe_Cor_Allcomp_Allvar2Select = dataframe_Cor_Allcomp_Allvar2[indice, ] varSelectTemp = dataframe_Cor_Allcomp_Allvar2Select$variable ind_Y = which(dataframe_Cor_Allcomp_Allvar2Select$bloc == "Y") if(length(ind_Y) != 0) { varSelect = varSelectTemp[- ind_Y] }else{ varSelect = varSelectTemp } matCor_Allcomp_AllvarSelect = matCor_Allcomp_Allvar[indice, , drop = FALSE] # Tracé de la superposition des cerles de corrélation. circle = list() circle[[1]] = ellipse(0, levels = 1, t = 1) circle[[2]] = ellipse(0, levels = 1, t = rad.in) circle = data.frame(do.call("rbind", circle), "Circle" = c(rep("Main circle", 100), rep("Inner circle", 100))) MainCircle = circle[grep("Main circle", circle[, 3]), ] InnerCircle = circle[grep("Inner circle", circle[, 3]), ] plot(MainCircle[, 1], MainCircle[, 2], type = "l", xlab = paste0("composante ", comp[1]), ylab = paste0("composante ", comp[2])) points(InnerCircle[, 1], InnerCircle[, 2], type = "l") if(dim(matCor_Allcomp_AllvarSelect)[1] != 0) { nom_blockEtReponseSelect = unique(dataframe_Cor_Allcomp_Allvar2Select$bloc) indice_blockEtReponseSelect = sapply(1:length(nom_blockEtReponseSelect), FUN = function(i){ ind = which(colnames(mat_cor_comp1) == nom_blockEtReponseSelect[i]) if(length(ind) != 0) { res = ind }else{ res = dim(mat_cor_comp1)[1] + 1 } return(res) }) vec_colSelect = vec_col[indice_blockEtReponseSelect] if(length(nom_blockEtReponseSelect) == 1) { points(matCor_Allcomp_AllvarSelect[, 1], matCor_Allcomp_AllvarSelect[, 2], col = NULL) text(matCor_Allcomp_AllvarSelect[, 1], matCor_Allcomp_AllvarSelect[, 2], labels = rownames(matCor_Allcomp_AllvarSelect), cex = cex, col = vec_colSelect[1]) }else{ nbVarSelect_bloc = cumsum(sapply(1:length(nom_blockEtReponseSelect), FUN = function(j){ res = length(which(dataframe_Cor_Allcomp_Allvar2Select$bloc == nom_blockEtReponseSelect[j])) return(res) })) for(i in 1:length(nbVarSelect_bloc)) { if(i == 1) { indice1 = 1:nbVarSelect_bloc[1] if(length(indice1) != 0) { matCor_Allcomp_AllvarSelect2 = matCor_Allcomp_AllvarSelect[indice1, , drop = FALSE] points(matCor_Allcomp_AllvarSelect2[, 1], matCor_Allcomp_AllvarSelect2[, 2], col = NULL) text(matCor_Allcomp_AllvarSelect2[, 1], matCor_Allcomp_AllvarSelect2[, 2], labels = rownames(matCor_Allcomp_AllvarSelect2), cex = cex, col = rep(vec_colSelect[i], dim(matCor_Allcomp_AllvarSelect2)[1])) }else{ cat(paste0("Il n'y a de variables dans cette zone du cercle de corrélation pour le bloc ", nom_blockEtReponseSelect[i]), "\n") } }else{ indice2 = (nbVarSelect_bloc[i - 1] + 1):nbVarSelect_bloc[i] if(length(indice2) != 0) { matCor_Allcomp_AllvarSelect2 = matCor_Allcomp_AllvarSelect[indice2, , drop = FALSE] points(matCor_Allcomp_AllvarSelect2[, 1], matCor_Allcomp_AllvarSelect2[, 2], col = NULL) text(matCor_Allcomp_AllvarSelect2[, 1], matCor_Allcomp_AllvarSelect2[, 2], labels = rownames(matCor_Allcomp_AllvarSelect2), cex = cex, col = rep(vec_colSelect[i], dim(matCor_Allcomp_AllvarSelect2)[1])) }else{ cat(paste0("Il n'y a de variables dans cette zone du cercle de corrélation pour le bloc ", nom_blockEtReponseSelect[i]), "\n") } } } } par(xpd = TRUE) legend(x = pos[1], y = pos[2], legend = nom_blockEtReponseSelect, pch = pch, col = vec_colSelect, cex = cex_legend, inset = inset) }else{ cat("Il n'y a de variables dans cette zone du cercle de corrélation", "\n") } } return(varSelect) } # La fonction networkVariable permet de tracer un réseau pour les variables de # certains blocs et une variable réponse. networkVariable <-function(dataframe_Cor_Allcomp_Allvar, vec_Var, nomVar_block_Y, comp = 1:2, cutoff = 0 ) { # Pour chaque variable de vec_Var et la variable réponse, nous récupérons les noms des blocs, # les corrélations entre la variable et chaque composante. ind = which(dataframe_Cor_Allcomp_Allvar$variable%in%vec_Var == TRUE) ind_Var_block_Y = which(dataframe_Cor_Allcomp_Allvar$variable == nomVar_block_Y) dataframe_Cor_Allcomp_AllvarSelect = dataframe_Cor_Allcomp_Allvar[c(ind, ind_Var_block_Y), ] nomBlocs = unique(dataframe_Cor_Allcomp_AllvarSelect$bloc) if((length(ind) == length(vec_Var)) & length(ind_Var_block_Y) == 1) { if(length(nomBlocs) == 1) { cat("Erreur : il n'y a des variables d'un seul bloc dans listeVar", "\n") }else{ liste_Cor_Allcomp_Allvar_bloc = lapply(1:length(nomBlocs), FUN = function(i){ nomBloc_i = nomBlocs[i] indice = which(dataframe_Cor_Allcomp_AllvarSelect$bloc == nomBloc_i) resTemp = dataframe_Cor_Allcomp_AllvarSelect[indice, ] if(length(indice) == 1) { res = resTemp }else{ res = resTemp } rownames(res) = resTemp$variable return(res) }) names(liste_Cor_Allcomp_Allvar_bloc) = nomBlocs coord = lapply(1:length(liste_Cor_Allcomp_Allvar_bloc), FUN = function(i){ resTemp = liste_Cor_Allcomp_Allvar_bloc[[i]] res = apply(resTemp[, comp + 2, drop = FALSE], 2, as.numeric) if(dim(resTemp)[1] == 1) { res = matrix(res, nrow = 1) } rownames(res) = resTemp$variable return(res) }) l = 1 M_block = list() node.X1 = node.X2 = w = NULL # Calcul de la similarité. for(j in 1:(length(nomBlocs) - 1)) { for(k in (j + 1):length(nomBlocs)) { M_block[[l]] = coord[[j]][, drop = FALSE] %*% t(coord[[k]][, drop = FALSE]) X1 = rownames(coord[[j]]) X2 = rownames(coord[[k]]) rep.X1 = rep(X1, each = length(X2)) rep.X2 = rep(X2, length(X1)) node.X1= c(node.X1, rep.X1) node.X2 = c(node.X2, rep.X2) w = c(w, as.vector(t(M_block[[l]]))) l = l + 1 } # Fin for(k in (j + 1):length(blocks)). } # Fin for(j in 1:(length(blocks) - 1)). # nodes group = NULL temp = lapply(1:length(liste_Cor_Allcomp_Allvar_bloc), function(i){ res = liste_Cor_Allcomp_Allvar_bloc[[i]]$variable return(res) }) names(temp) = names(liste_Cor_Allcomp_Allvar_bloc) for (i in 1:length(temp)) { group = c(group, rep(names(temp)[i], length(temp[[i]]))) } # Fin for (i in 1:length(temp)). nodes = data.frame(name = unlist(temp), group = group) # gR relations = data.frame(from = node.X1, to = node.X2, weight = w) idx = (abs(w) >= cutoff) relations = relations[idx, , drop = FALSE] gR = graph.data.frame(relations, directed = FALSE, vertices = nodes) gR = delete.vertices(gR, which(degree(gR) == 0)) res = list(gR = gR) l = 1 for (i in 1:(length(nomBlocs)-1)) { for (j in (i + 1):length(nomBlocs)) { res[paste("M", names(liste_Cor_Allcomp_Allvar_bloc)[i], names(liste_Cor_Allcomp_Allvar_bloc)[j], sep="_")] = list(M_block[[l]]) l = l + 1 } # Fin for (j in (i + 1):length(blocks)). } # Fin for (i in 1:(length(blocks)-1). res$cutoff = cutoff return(res) } # Fin if((length(which(ind))= length(listeVar)) & length(ind_Var_block_Y) == 1). }else{ if(length(ind) != length(vec_Var)) { cat("Erreur : les variables de vec_Var ne sont pas contenues dans dataframe_Cor_Allcomp_Allvar$variable.", "\n") } if(length(ind_Var_block_Y) == 1) { cat(paste0("Erreur : la variable réponse ", nomVar_block_Y," n'est pas contenue dans dataframe_Cor_Allcomp_Allvar$variable."), "\n") } } } # Integration 7 blocs norm gene ------------------------------------------- # La fonction compute_cor_comp_var calcule, pour chaque variable d'un bloc, les corrélations # entre la variable et les composantes sélectionnées par comp. compute_cor_comp_var <-function(res_block_splsda, comp = c(1:2)) { vec_blocksTemp = res_block_splsda$names$blocks ind_Y = which(vec_blocksTemp == "Y") vec_blocks1 = vec_blocksTemp[- ind_Y] vec_indice_blocks = sapply(1:length(vec_blocks1), FUN = function(i){ res = which(res_block_splsda$names$blocks == vec_blocks1[i]) return(res) }) # Calcul, pour chaque variable d'un bloc, des corrélations de cette variable avec # les composantes dont les indices sont indiqués dans le vecteur comp. vec_blocks2 = c() liste_cor_comp_var_global = list() for(i in 1:length(vec_indice_blocks)) { indice_blocks_i = vec_indice_blocks[i] block_i = res_block_splsda$names$blocks[indice_blocks_i] vec_varSelect_i = c() liste_comp_i = list() for(j in 1:length(comp)) { indice_comp_j = comp[j] liste_comp_i[[j]] = res_block_splsda$variates[[indice_blocks_i]][, indice_comp_j] vec_varSelect_comp_j = selectVar(res_block_splsda, comp = indice_comp_j)[[indice_blocks_i]][[1]] vec_varSelect_i = c(vec_varSelect_i, vec_varSelect_comp_j) } # Fin for(j in 1:length(comp)). vec_varSelect_i = unique(vec_varSelect_i) mat_block_i = res_block_splsda$X[[indice_blocks_i]][, vec_varSelect_i] liste_cor_comp_var = list() for(j in 1:length(liste_comp_i)) { liste_cor_comp_var[[j]] = sapply(1:dim(mat_block_i)[2], FUN = function(k){ cor(liste_comp_i[[j]], mat_block_i[, k]) }) } # Fin for(j in 1:length(liste_comp_i)). mat_cor_Allcomp_Allvar = Reduce(cbind, liste_cor_comp_var) rownames(mat_cor_Allcomp_Allvar) = colnames(mat_block_i) vec_blocks_i = rep(block_i, dim(mat_block_i)[2]) vec_blocks2 = c(vec_blocks2, vec_blocks_i) liste_cor_comp_var_global[[i]] = mat_cor_Allcomp_Allvar } # Fin for(i in 1:length(blocks)). mat_cor_comp_var_global = Reduce(rbind, liste_cor_comp_var_global) dataframe_cor_comp_var_global = data.frame(cbind(rownames(mat_cor_comp_var_global), vec_blocks2, mat_cor_comp_var_global)) colnames(dataframe_cor_comp_var_global) = c("variable", "bloc", paste0("cor_var_comp", comp)) dataframe_cor_comp_var_global[, 1:2] = apply(dataframe_cor_comp_var_global[, 1:2], 2, as.character) dataframe_cor_comp_var_global[, 3:dim(dataframe_cor_comp_var_global)[2]] = apply(dataframe_cor_comp_var_global[, 3:dim(dataframe_cor_comp_var_global)[2]], 2, as.numeric) return(dataframe_cor_comp_var_global) } # La fonction composanteColin renvoie une liste. Le ième élément de cette liste # contient les indices des blocs tels que, pour chaque paire de ces blocs, la # première composante du bloc 1 est fortement corrélée à la première composante # du bloc2 en valeur absolue et la deuxième composante du bloc 1 est fortement # corrélée à la deuxième composante du bloc2 en valeur absolue. composanteColin <-function(mat_cor_comp1, mat_cor_comp2, cutoff_comp) { res = list() index = 1:dim(mat_cor_comp1)[1] i = 1 compt = 1 while(length(index) != 0 & compt <= dim(mat_cor_comp1)[1]) { index_i = index[i] res[[compt]] = c(index_i) index2 = index[-i] if(length(index2) != 1) { for(j in index2) { if(length(res[[compt]]) == 1) { if(abs(mat_cor_comp1[j, index_i]) > cutoff_comp & abs(mat_cor_comp2[j, index_i]) > cutoff_comp) { res[[compt]] = c(res[[compt]], j) index = index[- c(which(index == j))] } }else{ indice = sapply(1:length(res[[compt]]), FUN = function(k){ index_k = res[[compt]][k] cond = abs(mat_cor_comp1[index_k, j]) > cutoff_comp & abs(mat_cor_comp2[index_k, j]) > cutoff_comp return(cond) }) if(all(indice)) { res[[compt]] = c(res[[compt]], j) index = index[- c(which(index == j))] } } } }else{ res[[compt]] = index_i } index = index[- c(which(index == index_i))] compt = compt + 1 } # Fin for(i in 1:dim(mat_cor_comp1)[2]). if(length(index) != 0) { for(i in 1:length(index)) { res = c(res, list(index[i])) } } return(res) } # La fonction compute_blockSelect permet de déterminer toutes les combinaisons # possibles des blocs qui peuvent être superposés dans le cercle de corrélations # et dont les variables peuvent être présentes dans le réseau. compute_blockSelect <-function(mat_cor_comp1, mat_cor_comp2, cutoff_comp) { liste_vec_indice_blockSelect = composanteColin(mat_cor_comp1, mat_cor_comp2, cutoff_comp) vec_blocks = colnames(mat_cor_comp1) liste_vec_blocks = list() # liste_blocks est une liste contenant toutes les combinaisons possibles de blocs dont # on peut superposer les cercles de corrélations et dont les variables peuvent être # présentes dans le réseau. for(i in 1:length(liste_vec_indice_blockSelect)) { vec_indice_blockSelect_i = liste_vec_indice_blockSelect[[i]] for(k in 1:length(vec_indice_blockSelect_i)) { matComb = combn(vec_indice_blockSelect_i, m = k) liste_vec_blocks_i = lapply(1:dim(matComb)[2], FUN = function(i){ comb_i = matComb[, i] resultat = vec_blocks[comb_i] return(resultat) }) for(j in 1:length(liste_vec_blocks_i)) { liste_vec_blocks = c(liste_vec_blocks, list(liste_vec_blocks_i[[j]])) } # Fin for(j in 1:length(liste_noms_blocks)). } # Fin for(k in 1:length(blockSelect_i)). } # Fin for(i in 1:length(blockSelect)). liste_vec_blocks = unique(liste_vec_blocks) return(list(liste_vec_indice_blockSelect = liste_vec_indice_blockSelect, liste_vec_blocks = liste_vec_blocks)) } # La fonction matCorEtBlockSelect permet de calculer la matrice de # corrélation entre les comp[1] composantes de chaque bloc et la # matrice de corrélation entre les comp[2] composantes de chaque # bloc. Elle permet aussi de calculer, pour chaque variable d'un bloc, # les corrélations entre les composantes comp[1] et comp[2] et cette variable # et de calculer toutes les combinaisons possibles des blocs pour lesquels nous # pouvons superposer les cercles de corrélations. matCorEtBlockSelect <-function(res_block_splsda, cutoff_comp, comp) { liste_mat_cor_comp = matriceCorrelation_comp(res_block_splsda = res_block_splsda, comp = comp) mat_cor_comp1 = liste_mat_cor_comp$mat_cor_comp1 mat_cor_comp2 = liste_mat_cor_comp$mat_cor_comp2 dataframe_cor_comp_var_global = compute_cor_comp_var(res_block_splsda = res_block_splsda, comp = comp) liste_blockSelect = compute_blockSelect(mat_cor_comp1 = mat_cor_comp1, mat_cor_comp2 = mat_cor_comp2, cutoff_comp = cutoff_comp) liste_vec_indice_blockSelect = liste_blockSelect$liste_vec_indice_blockSelect liste_vec_blocks = liste_blockSelect$liste_vec_blocks return(list(mat_cor_comp1 = mat_cor_comp1, mat_cor_comp2 = mat_cor_comp2, dataframe_cor_comp_var_global = dataframe_cor_comp_var_global, liste_vec_indice_blockSelect = liste_vec_indice_blockSelect, liste_vec_blocks = liste_vec_blocks)) } # La fonction addVariablesReponses permet de calculer, pour chaque # variable réponse, la corrélation entre cette variable et la # comp[1] composante et la corrélation entre cette variable et la comp[2] # composante pour chaque groupe de blocs. addVariablesReponses <-function(res_block_splsda, dataframe_cor_comp_var_global, liste_vec_indice_blockSelect, mat_block_Y) { # On récupère les indices des composantes utilisées pour calculer les # corrélations entre les variables des blocs et les composantes. comp = as.numeric(sapply(3:4, FUN = function(i){ col_i = colnames(dataframe_cor_comp_var_global)[i] ch = strsplit(col_i, split = "_")[[1]] resTemp = ch[length(ch)] res = substring(resTemp, nchar(resTemp), nchar(resTemp)) return(res) })) # On calcule, pour chaque variable réponse, pour chaque composante du premier # bloc du groupe de blocs, la corrélation entre cette variable réponse et la composante. liste_dataframe_cor_allcomp_varReponses = list() vec_groupe_blocks = c() for(i in 1:length(liste_vec_indice_blockSelect)) { vec_indice_blockSelect_i = liste_vec_indice_blockSelect[[i]] indice_first_block_i = vec_indice_blockSelect_i[1] liste_comp_i = list() liste_cor_comp_var = list() for(j in 1:length(comp)) { indice_comp_j = comp[j] comp_j = res_block_splsda$variates[[indice_first_block_i]][, indice_comp_j] liste_cor_comp_var[[j]] = sapply(1:dim(mat_block_Y)[2], FUN = function(k){ cor(comp_j, mat_block_Y[, k]) }) } # Fin for(j in 1:length(liste_comp_i)). mat_cor_allcomp_varReponses = Reduce(cbind, liste_cor_comp_var) rownames(mat_cor_allcomp_varReponses) = colnames(mat_block_Y) dataframe_allcomp_varReponses = data.frame(colnames(mat_block_Y), rep("Y", dim(mat_block_Y)[2]), mat_cor_allcomp_varReponses) colnames(dataframe_allcomp_varReponses) = c("variable", "bloc", paste0("cor_var_comp", comp)) liste_dataframe_cor_allcomp_varReponses[[i]] = dataframe_allcomp_varReponses groupe_blocks_i = res_block_splsda$names$blocks[vec_indice_blockSelect_i] vec_groupe_blocks = c(vec_groupe_blocks, paste(groupe_blocks_i, collapse = "-")) } # Fin for(in in 1:length(blockSelect)). names(liste_dataframe_cor_allcomp_varReponses) = vec_groupe_blocks liste_dataframe_cor_comp_var_global = list() for(i in 1:length(liste_vec_indice_blockSelect)) { vec_indice_blockSelect_i = liste_vec_indice_blockSelect[[i]] groupe_blocks_i = res_block_splsda$names$blocks[vec_indice_blockSelect_i] indice_i = which(dataframe_cor_comp_var_global$bloc%in%groupe_blocks_i == TRUE) dataframe_cor_comp_var_global_indice_i = dataframe_cor_comp_var_global[indice_i, ] dataframe_cor_comp_varBlockEtVarRep_global = rbind(dataframe_cor_comp_var_global_indice_i, liste_dataframe_cor_allcomp_varReponses[[i]]) dataframe_cor_comp_varBlockEtVarRep_global[, 1:2] = apply(dataframe_cor_comp_varBlockEtVarRep_global[, 1:2], 2, as.character) dataframe_cor_comp_varBlockEtVarRep_global[, 3:dim(dataframe_cor_comp_varBlockEtVarRep_global)[2]] = apply(dataframe_cor_comp_varBlockEtVarRep_global[, 3:dim(dataframe_cor_comp_varBlockEtVarRep_global)[2]], 2, as.numeric) liste_dataframe_cor_comp_var_global[[i]] = dataframe_cor_comp_varBlockEtVarRep_global } # Fin for(i in 1:length(indice_blockSelect)). names(liste_dataframe_cor_comp_var_global) = vec_groupe_blocks return(liste_dataframe_cor_comp_var_global) } # La fonction addVariablesReponsesModified permet de calculer, pour chaque # variable réponse, la corrélation entre cette variable et la # comp[1] composante et la corrélation entre cette variable et la comp[2] # composante pour chaque groupe de blocs. addVariablesReponsesModified <-function(res_block_splsda, dataframe_cor_comp_var_global, liste_vec_indice_blockSelect, mat_block_Y) { # On récupère les indices des composantes utilisées pour calculer les # corrélations entre les variables des blocs et les composantes. comp = as.numeric(sapply(3:4, FUN = function(i){ col_i = colnames(dataframe_cor_comp_var_global)[i] ch = strsplit(col_i, split = "_")[[1]] resTemp = ch[length(ch)] res = substring(resTemp, nchar(resTemp), nchar(resTemp)) return(res) })) # On calcule, pour chaque variable réponse, pour chaque composante du premier # bloc du groupe de blocs, la corrélation entre cette variable réponse et la composante. liste_dataframe_cor_allcomp_varReponses = list() vec_groupe_blocks = c() for(i in 1:length(liste_vec_indice_blockSelect)) { vec_indice_blockSelect_i = liste_vec_indice_blockSelect[[i]] indice_first_block_i = vec_indice_blockSelect_i[1] liste_comp_i = list() liste_cor_comp_var = list() for(j in 1:length(comp)) { indice_comp_j = comp[j] comp_j = res_block_splsda$variates[[indice_first_block_i]][, indice_comp_j] liste_cor_comp_var[[j]] = sapply(1:dim(mat_block_Y)[2], FUN = function(k){ resTemp = comp_j%*%mat_block_Y[, k] res = resTemp/(norm(comp_j - mean(comp_j), "2")*norm(mat_block_Y[, k], "2")) }) } # Fin for(j in 1:length(liste_comp_i)). mat_cor_allcomp_varReponses = Reduce(cbind, liste_cor_comp_var) rownames(mat_cor_allcomp_varReponses) = colnames(mat_block_Y) dataframe_allcomp_varReponses = data.frame(colnames(mat_block_Y), rep("Y", dim(mat_block_Y)[2]), mat_cor_allcomp_varReponses) colnames(dataframe_allcomp_varReponses) = c("variable", "bloc", paste0("cor_var_comp", comp)) liste_dataframe_cor_allcomp_varReponses[[i]] = dataframe_allcomp_varReponses groupe_blocks_i = res_block_splsda$names$blocks[vec_indice_blockSelect_i] vec_groupe_blocks = c(vec_groupe_blocks, paste(groupe_blocks_i, collapse = "-")) } # Fin for(in in 1:length(blockSelect)). names(liste_dataframe_cor_allcomp_varReponses) = vec_groupe_blocks liste_dataframe_cor_comp_var_global = list() for(i in 1:length(liste_vec_indice_blockSelect)) { vec_indice_blockSelect_i = liste_vec_indice_blockSelect[[i]] groupe_blocks_i = res_block_splsda$names$blocks[vec_indice_blockSelect_i] indice_i = which(dataframe_cor_comp_var_global$bloc%in%groupe_blocks_i == TRUE) dataframe_cor_comp_var_global_indice_i = dataframe_cor_comp_var_global[indice_i, ] dataframe_cor_comp_varBlockEtVarRep_global = rbind(dataframe_cor_comp_var_global_indice_i, liste_dataframe_cor_allcomp_varReponses[[i]]) dataframe_cor_comp_varBlockEtVarRep_global[, 1:2] = apply(dataframe_cor_comp_varBlockEtVarRep_global[, 1:2], 2, as.character) dataframe_cor_comp_varBlockEtVarRep_global[, 3:dim(dataframe_cor_comp_varBlockEtVarRep_global)[2]] = apply(dataframe_cor_comp_varBlockEtVarRep_global[, 3:dim(dataframe_cor_comp_varBlockEtVarRep_global)[2]], 2, as.numeric) liste_dataframe_cor_comp_var_global[[i]] = dataframe_cor_comp_varBlockEtVarRep_global } # Fin for(i in 1:length(indice_blockSelect)). names(liste_dataframe_cor_comp_var_global) = vec_groupe_blocks return(liste_dataframe_cor_comp_var_global) } # La fonction circleCor permet de zoomer sur un rectangle du cercle de corrélations # et de récupérer les variables des blocs dans cette partie zoomée du cercle de # corrélations. circleCor <-function(liste_dataframe_cor_comp_var_global, liste_vec_indice_blockSelect, mat_cor_comp1, mat_cor_comp2, vec_blocks, nomsVarReponses, cutoff = 0.85, min.X = -1, max.X = 1, min.Y = -1, max.Y = 1, vec_col = colorRampPalette(brewer.pal(9, "Spectral"))(dim(mat_cor_comp1)[1] + 1), rad.in = 0.5, cex = 0.7, cex_legend = 0.8, pos = c(1.2, 0), pch = 20, inset = c(-0.25, 0)) { # On vérifie que nous pouvons superposer les cercles de corrélation. vec_indice_blocks = sapply(1:length(vec_blocks), FUN = function(i){ res = which(colnames(mat_cor_comp1) == vec_blocks[i]) return(res) }) boolean = FALSE i = 1 while(i <= length(liste_vec_indice_blockSelect) & !boolean) { vec_indice_blockSelect_i = liste_vec_indice_blockSelect[[i]] cond = length(which(vec_indice_blockSelect_i%in%vec_indice_blocks == TRUE)) == length(vec_blocks) if(cond) { boolean = TRUE } i = i + 1 } # Fin while(i <= length(liste_vec_indice_blockSelect) & !boolean). varSelect = NULL if(!boolean) { stop(paste0("The blocks : ", paste(vec_blocks, collapse = ", "), " can not be superimposed.")) }else{ # On récupère le groupe de blocs auxquels appartient vec_blocks. indice_nomsBlock = sapply(1:length(liste_dataframe_cor_comp_var_global), FUN = function(i){ name_iTemp = names(liste_dataframe_cor_comp_var_global)[i] name_i = strsplit(name_iTemp, split = "-")[[1]] res = all(vec_blocks%in%name_i) return(res) }) dataframe_cor_comp_var_global = liste_dataframe_cor_comp_var_global[[which(indice_nomsBlock == TRUE)]] dataframe_cor_comp_var_globalTemp1 = dataframe_cor_comp_var_global indice_nomsVarReponses = all(nomsVarReponses%in%dataframe_cor_comp_var_globalTemp1$variable) if(!indice_nomsVarReponses) { stop("All the correlations between the response variables and the first component and the correlations between the responses variables and the second component have not been computed.") }else{ comp = as.numeric(sapply(3:4, FUN = function(i){ col_i = colnames(dataframe_cor_comp_var_globalTemp1)[i] ch = strsplit(col_i, split = "_")[[1]] resTemp = ch[length(ch)] res = substring(resTemp, nchar(resTemp), nchar(resTemp)) return(res) })) mat_cor_comp_var_globalTemp1 = t(sapply(1:dim(dataframe_cor_comp_var_globalTemp1)[1], FUN = function(i){ dataframe_cor_comp_var_globalTemp1_i = dataframe_cor_comp_var_globalTemp1[i, ] block_i = dataframe_cor_comp_var_globalTemp1_i$bloc cor1 = dataframe_cor_comp_var_globalTemp1_i[paste0("cor_var_comp", comp[1])] cor2 = dataframe_cor_comp_var_globalTemp1_i[paste0("cor_var_comp", comp[2])] if(block_i == "Y") { cor1_sign = cor1 cor2_sign = cor2 }else{ indice_block_comp1_i = which(colnames(mat_cor_comp1) == block_i) indice_block_comp2_i = which(colnames(mat_cor_comp2) == block_i) cor1_sign = sign(mat_cor_comp1[vec_indice_blocks[1], indice_block_comp1_i])*cor1 cor2_sign = sign(mat_cor_comp2[vec_indice_blocks[1], indice_block_comp2_i])*cor2 } res = c(dataframe_cor_comp_var_globalTemp1_i[1:2], cor1_sign, cor2_sign) return(res) })) dataframe_cor_comp_var_globalTemp2 = as.data.frame(mat_cor_comp_var_globalTemp1) dataframe_cor_comp_var_globalTemp2[, 1:2] = apply(dataframe_cor_comp_var_globalTemp2[, 1:2], 2, as.character) colnames(dataframe_cor_comp_var_globalTemp2) = colnames(dataframe_cor_comp_var_global) # Pour les variables de vec_blocks et les variables réponses nomsVarReponses, on récupère les corrélations # entre ces variables et les composantes. indice1 = sapply(1:dim(dataframe_cor_comp_var_globalTemp2)[1], FUN = function(i){ block_i = dataframe_cor_comp_var_globalTemp2$bloc[i] if(block_i == "Y") { res = dataframe_cor_comp_var_globalTemp2$variable[i]%in%nomsVarReponses }else{ res = block_i%in%vec_blocks } return(res) }) dataframe_cor_comp_var_global2 = dataframe_cor_comp_var_globalTemp2[indice1, ] mat_cor_comp_var_global2 = apply(dataframe_cor_comp_var_global2[ , 3:4], 2, as.numeric) rownames(mat_cor_comp_var_global2) = dataframe_cor_comp_var_global2$variable # indice permet de récupérer les variables de chaque bloc fortement corrélées avec soit # la première composante ou la deuxième composante dans un rectangle du cercle de corrélations # et de récupérer les variables réponses. indice2 = sapply(1:dim(dataframe_cor_comp_var_global2)[1], FUN = function(k){ cor1 = mat_cor_comp_var_global2[k, 1] cor2 = mat_cor_comp_var_global2[k, 2] blockEtReponse_k = dataframe_cor_comp_var_global2[k, 2] if(blockEtReponse_k == "Y") { cond2 = TRUE }else{ cond1 = abs(cor1) > cutoff | abs(cor2) > cutoff cond2 = cor1 > min.X & cor1 < max.X & cor2 > min.Y & cor2 < max.Y & cond1 } return(cond2) }) dataframe_cor_comp_var_global2Select = dataframe_cor_comp_var_global2[indice2, ] varSelectTemp = dataframe_cor_comp_var_global2Select$variable ind_Y = which(dataframe_cor_comp_var_global2Select$bloc == "Y") if(length(ind_Y) != 0) { varSelect = varSelectTemp[- ind_Y] }else{ varSelect = varSelectTemp } mat_cor_comp_var_global2Select = mat_cor_comp_var_global2[indice2, , drop = FALSE] # Tracé de la superposition des cerles de corrélation. circle = list() circle[[1]] = ellipse(0, levels = 1, t = 1) circle[[2]] = ellipse(0, levels = 1, t = rad.in) circle = data.frame(do.call("rbind", circle), "Circle" = c(rep("Main circle", 100), rep("Inner circle", 100))) MainCircle = circle[grep("Main circle", circle[, 3]), ] InnerCircle = circle[grep("Inner circle", circle[, 3]), ] plot(MainCircle[, 1], MainCircle[, 2], type = "l", xlab = paste0("composante ", comp[1]), ylab = paste0("composante ", comp[2])) points(InnerCircle[, 1], InnerCircle[, 2], type = "l") if(dim(mat_cor_comp_var_global2Select)[1] != 0) { vec_blockEtReponseSelect = unique(dataframe_cor_comp_var_global2Select$bloc) indice_blockEtReponseSelect = sapply(1:length(vec_blockEtReponseSelect), FUN = function(i){ ind = which(colnames(mat_cor_comp1) == vec_blockEtReponseSelect[i]) if(length(ind) != 0) { res = ind }else{ res = dim(mat_cor_comp1)[1] + 1 } return(res) }) vec_colSelect = vec_col[indice_blockEtReponseSelect] if(length(vec_blockEtReponseSelect) == 1) { points(mat_cor_comp_var_global2Select[, 1], mat_cor_comp_var_global2Select[, 2], col = NULL) text(mat_cor_comp_var_global2Select[, 1], mat_cor_comp_var_global2Select[, 2], labels = rownames(mat_cor_comp_var_global2Select), cex = cex, col = vec_colSelect[1]) }else{ nbVarSelect_bloc = cumsum(sapply(1:length(vec_blockEtReponseSelect), FUN = function(j){ res = length(which(dataframe_cor_comp_var_global2Select$bloc == vec_blockEtReponseSelect[j])) return(res) })) for(i in 1:length(nbVarSelect_bloc)) { if(i == 1) { indice_nbVar1 = 1:nbVarSelect_bloc[1] mat_cor_comp_var_global2Select2 = mat_cor_comp_var_global2Select[indice_nbVar1, , drop = FALSE] points(mat_cor_comp_var_global2Select2[, 1], mat_cor_comp_var_global2Select2[, 2], col = NULL) text(mat_cor_comp_var_global2Select2[, 1], mat_cor_comp_var_global2Select2[, 2], labels = rownames(mat_cor_comp_var_global2Select2), cex = cex, col = rep(vec_colSelect[i], dim(mat_cor_comp_var_global2Select2)[1])) }else{ indice_nbVar2 = (nbVarSelect_bloc[i - 1] + 1):nbVarSelect_bloc[i] mat_cor_comp_var_global2Select2 = mat_cor_comp_var_global2Select[indice_nbVar2, , drop = FALSE] points(mat_cor_comp_var_global2Select2[, 1], mat_cor_comp_var_global2Select2[, 2], col = NULL) text(mat_cor_comp_var_global2Select2[, 1], mat_cor_comp_var_global2Select2[, 2], labels = rownames(mat_cor_comp_var_global2Select2), cex = cex, col = rep(vec_colSelect[i], dim(mat_cor_comp_var_global2Select2)[1])) } } } par(xpd = TRUE) legend(x = pos[1], y = pos[2], legend = vec_blockEtReponseSelect, pch = pch, col = vec_colSelect, cex = cex_legend, inset = inset) }else{ warning("There is no variables in this rectangle of the correlation circle.") } } } return(varSelect) } tabVarSelect <-function(varSelect) { } # La fonction compute_matSimilarity calcule, pour chaque groupe de blocs, les # similarités entre le bloc1 et le bloc2 et entre les blocs et la réponse. compute_matSimilarity <-function(liste_dataframe_cor_comp_var_global) { comp = as.numeric(sapply(3:4, FUN = function(i){ col_i = colnames(liste_dataframe_cor_comp_var_global[[1]])[i] ch = strsplit(col_i, split = "_")[[1]] resTemp = ch[length(ch)] res = substring(resTemp, nchar(resTemp), nchar(resTemp)) return(res) })) liste_matSimilarity_group = list() # On calcule, pour chaque groupe de blocs, les matrices de similarités entre # chaque paire de blocs. for(i in 1:length(liste_dataframe_cor_comp_var_global)) { dataframe_cor_comp_var_global_i = liste_dataframe_cor_comp_var_global[[i]] blocks_i = unique(dataframe_cor_comp_var_global_i$bloc) coord = lapply(1:length(blocks_i), FUN = function(j){ ind_j = which(dataframe_cor_comp_var_global_i$bloc == blocks_i[j]) dataframe_cor_comp_var_global_i_j = dataframe_cor_comp_var_global_i[ind_j, paste0("cor_var_comp", comp)] res = as.matrix(dataframe_cor_comp_var_global_i_j, drop = FALSE) rownames(res) = rownames(dataframe_cor_comp_var_global_i_j) return(res) }) M_block = list() l = 1 vec_blocks_i = c() for(j in 1:(length(blocks_i) - 1)) { blocks_i_j = blocks_i[j] for(k in (j + 1):length(blocks_i)) { blocks_i_k = blocks_i[k] M_block[[l]] = coord[[j]][, drop = FALSE] %*% t(coord[[k]][, drop = FALSE]) rownames(M_block[[l]]) = rownames(coord[[j]]) colnames(M_block[[l]]) = rownames(coord[[k]]) blocks_j_k = paste(c(blocks_i_j, blocks_i_k), collapse = "-") vec_blocks_i = c(vec_blocks_i, blocks_j_k) l = l + 1 } # Fin for(k in (j + 1):length(blocks)). } # Fin for(j in 1:(length(blocks) - 1)). names(M_block) = vec_blocks_i liste_matSimilarity_group[[i]] = M_block } # Fin for(i in 1:length(liste_dataframe_Cor_comp_var_global)). names(liste_matSimilarity_group) = names(liste_dataframe_cor_comp_var_global) return(list(liste_matSimilarity_group = liste_matSimilarity_group, comp = comp)) } # La fonction networkVariableSelect permet de tracer un réseau pour les variables de # certains blocs et des variables réponses. networkVariableSelect <-function(liste_matSimilarity_group, comp, res_block_splsda, cutoff_comp = 0.8, vec_varBlock, vec_varRep, cutoff = 0 ) { vec_varBlockEtReponse = c(vec_varBlock, vec_varRep) # Nous vérifions que nous pouvons créer un réseau pour les variables des # blocs vec_Var_blockEtReponse. # Nous recherchons le groupe de blocs associé à vec_varBlock. indice_group_vecVar = sapply(1:length(liste_matSimilarity_group), FUN = function(i){ liste_matSimilarity_group_i = liste_matSimilarity_group[[i]] boolean = FALSE j = 1 while((j <= length(liste_matSimilarity_group_i))&!boolean) { matSimilarity_group_i_j = liste_matSimilarity_group_i[[j]] vec_var_block1 = rownames(matSimilarity_group_i_j) vec_var_block2 = colnames(matSimilarity_group_i_j) vec_var_block = c(vec_var_block1, vec_var_block2) if(any(vec_var_block%in%vec_varBlock)) { boolean = TRUE } j = j + 1 } # Fin while((j <= length(liste_matSimilarity_group_i))&!boolean). res = boolean return(res) }) res = NULL if(length(which(indice_group_vecVar == TRUE)) >= 2) { cat("Erreur : les variables de vec_varBlock. doivent appartenir à un seul élément de la liste liste_res_matSimilarity_group$liste_matSimilarity_group.", "\n") }else{ liste_matSimilarity = liste_matSimilarity_group[[which(indice_group_vecVar == TRUE)]] blocks_liste_matSimilarityTemp1 = sapply(1:length(liste_matSimilarity), FUN = function(i){ noms_block1_block2_i = names(liste_matSimilarity)[i] ch = strsplit(noms_block1_block2_i, split = "-")[[1]] block1 = ch[1] block2 = ch[2] res = c(block1, block2) return(res) }) blocks_liste_matSimilarity = unique(as.vector(blocks_liste_matSimilarityTemp1)) indice_blocks_liste_matSimilarityTemp = sapply(1:length(blocks_liste_matSimilarity), FUN = function(i){ res = which(res_block_splsda$names$blocks == blocks_liste_matSimilarity[i]) return(res) }) ind_Y = which(res_block_splsda$names$blocks == "Y") indice_blocks_liste_matSimilarity = indice_blocks_liste_matSimilarityTemp[indice_blocks_liste_matSimilarityTemp != ind_Y] boolean_pos_cor = TRUE # Nous vérifions que les ièmes composantes de chaque bloc sont fortement corrélées positivement. if(length(indice_blocks_liste_matSimilarity) == 1) { }else{ for(i in 1:length(comp)) { comp_i = comp[i] for(j in 1:(length(indice_blocks_liste_matSimilarity) - 1)) { indice_blocks_liste_matSimilarity_j = indice_blocks_liste_matSimilarity[j] comp_indice_blocks_liste_matSimilarity_j = res_block_splsda$variates[[indice_blocks_liste_matSimilarity_j]][, comp_i] for(k in (j + 1):length(indice_blocks_liste_matSimilarity)) { indice_blocks_liste_matSimilarity_k = indice_blocks_liste_matSimilarity[k] comp_indice_blocks_liste_matSimilarity_k = res_block_splsda$variates[[indice_blocks_liste_matSimilarity_k]][, comp_i] cor = cor(comp_indice_blocks_liste_matSimilarity_j, comp_indice_blocks_liste_matSimilarity_k) boolean_pos_cor = boolean_pos_cor & all(cor > cutoff_comp) } # Fin for(k in (j + 1):length(indice_blocks_liste_matSimilarity)). } # Fin for(j in 1:(length(indice_blocks_liste_matSimilarity) - 1)). } # Fin for(i in 1:length(comp)). } if(!boolean_pos_cor) { cat("Erreur : pour chaque paire de bloc, la ième composante de chaque bloc doivent être corrélées positivement afin de pouvoir créer un réseau.", "\n") }else{ # Nous récupérons les matrices de similarités associés aux variables de vec_varBlockEtReponse . liste_matSimilaritySelectTemp = lapply(1:length(liste_matSimilarity), FUN = function(j){ matSimilarity_j = liste_matSimilarity[[j]] indice_row_matSimilarity_j = which(rownames(matSimilarity_j)%in%vec_varBlockEtReponse == TRUE) indice_col_matSimilarity_j = which(colnames(matSimilarity_j)%in%vec_varBlockEtReponse == TRUE) if((length(indice_row_matSimilarity_j) != 0) &(length(indice_col_matSimilarity_j) != 0)) { res = matSimilarity_j[indice_row_matSimilarity_j, indice_col_matSimilarity_j, drop = FALSE] }else{ res = NA } return(res) }) names(liste_matSimilaritySelectTemp) = names(liste_matSimilarity) indice_NA_liste_matSimilaritySelectTemp = sapply(1:length(liste_matSimilaritySelectTemp), FUN = function(i){ liste_matSimilaritySelectTemp_i = liste_matSimilaritySelectTemp[[i]] if(is.matrix(liste_matSimilaritySelectTemp_i)) { res = FALSE }else{ if(is.na(liste_matSimilaritySelectTemp_i)) { res = TRUE }else{ res = FALSE } } return(res) }) liste_matSimilaritySelect = liste_matSimilaritySelectTemp[!indice_NA_liste_matSimilaritySelectTemp] # Nous créons le réseau. w = c() node.X1 = c() node.X2 = c() vec_group = c() vec_nomsVar = c() for(i in 1:length(liste_matSimilaritySelect)) { noms_bloc1_bloc2 = names(liste_matSimilaritySelect)[i] matSimilaritySelect_i = liste_matSimilaritySelect[[i]] X1 = rownames(matSimilaritySelect_i) X2 = colnames(matSimilaritySelect_i) rep.X1 = rep(X1, each = length(X2)) rep.X2 = rep(X2, length(X1)) node.X1= c(node.X1, rep.X1) node.X2 = c(node.X2, rep.X2) ch = strsplit(noms_bloc1_bloc2, split = "-")[[1]] nom_bloc1 = ch[1] nom_bloc2 = ch[2] vec_group = c(vec_group, c(rep(nom_bloc1, length(X1)), rep(nom_bloc2, length(X2)))) vec_nomsVar = c(vec_nomsVar, c(X1, X2)) w = c(w, as.vector(t(matSimilaritySelect_i))) } # Fin for(i in 1:length(liste_matSimilaritySelect)). dup = duplicated(vec_nomsVar) vec_nomsVar = vec_nomsVar[!dup] vec_group = vec_group[!dup] nodes = data.frame(name = vec_nomsVar, group = vec_group) # gR relations = data.frame(from = node.X1, to = node.X2, weight = w) idx = (abs(w) >= cutoff) relations = relations[idx, , drop = FALSE] gR = graph.data.frame(relations, directed = FALSE, vertices = nodes) # On supprime les noeuds qui n'ont pas d'arêtes. gR = delete.vertices(gR, which(degree(gR) == 0)) res = list(gR = gR) res$cutoff = cutoff } } return(res) } # La fonction networkVar permet de tracer un réseau pour certaines variables des # blocs et des variables réponses. networkVar <-function(liste_matSimilarity_group = liste_matSimilarity_group, comp = comp, res_block_splsda, cutoff_comp = 0.8, vec_varBlock, vec_varRep, vec_varBlockInteret = NULL, cutoff = 0 ) { # Nous vérifions que nous pouvons créer un réseau pour les variables des # blocs vec_VarBlock. # Nous recherchons le groupe de blocs associé à vec_Var. indice_group_vecVar = sapply(1:length(liste_matSimilarity_group), FUN = function(i){ liste_matSimilarity_group_i = liste_matSimilarity_group[[i]] boolean = FALSE j = 1 while((j <= length(liste_matSimilarity_group_i))&!boolean) { matSimilarity_group_i_j = liste_matSimilarity_group_i[[j]] vec_var_block1 = rownames(matSimilarity_group_i_j) vec_var_block2 = colnames(matSimilarity_group_i_j) vec_var_block1_block2 = c(vec_var_block1, vec_var_block2) if(any(vec_var_block1_block2%in%vec_varBlock)) { boolean = TRUE } j = j + 1 } # Fin while((j <= length(liste_matSimilarity_group_i))&!boolean). res = boolean return(res) }) if(length(which(indice_group_vecVar == TRUE)) >= 2) { stop("The variables of vec_var have to belong to only one element of liste_res_matSimilarity_group$liste_matSimilarity_group.") }else{ liste_matSimilarity = liste_matSimilarity_group[[which(indice_group_vecVar == TRUE)]] if(!is.null(vec_varBlockInteret)) { AllVariables_vec = sapply(1:length(liste_matSimilarity), FUN = function(i){ matSimilarity_i = liste_matSimilarity[[i]] res = c(rownames(matSimilarity_i), colnames(matSimilarity_i)) return(res) }) AllVariables_vec = unique(unlist(AllVariables_vec)) index_variableInterestNotInAllVariables_vec = vec_varBlockInteret%in%AllVariables_vec if(length(which(index_variableInterestNotInAllVariables_vec == FALSE)) != 0) { InterestVariableNotIn = vec_varBlockInteret[which(index_variableInterestNotInAllVariables_vec == FALSE)] warning(paste0("The variables of interest ", paste(InterestVariableNotIn, collapse = ","), " do not belong to the variables of the blocks for which the network can be created. These variables will be not in the network.")) } vec_varBlockInteret2 = vec_varBlockInteret[which(index_variableInterestNotInAllVariables_vec == TRUE)] }else{ vec_varBlockInteret2 = vec_varBlockInteret } if(!is.null(vec_varBlock) & !is.null(vec_varBlockInteret2)) { varCom = intersect(vec_varBlock, vec_varBlockInteret2) if(length(varCom) != 0) { index_varCom = sapply(1:length(varCom), FUN = function(i){ res = which(vec_varBlock == varCom[i]) return(res) }) vec_varBlock2 = vec_varBlock[- index_varCom] }else{ vec_varBlock2 = vec_varBlock } vec_varBlock3 = c(vec_varBlock2, vec_varBlockInteret2) }else if(!is.null(vec_varBlock) & is.null(vec_varBlockInteret2)) { vec_varBlock3 = vec_varBlock }else if(is.null(vec_varBlock) & !is.null(vec_varBlockInteret2)) { vec_varBlock3 = vec_varBlockInteret2 } vec_var = c(vec_varBlock3, vec_varRep) blocks_liste_matSimilarityTemp1 = sapply(1:length(liste_matSimilarity), FUN = function(i){ noms_block1_block2_i = names(liste_matSimilarity)[i] ch = strsplit(noms_block1_block2_i, split = "-")[[1]] block1 = ch[1] block2 = ch[2] res = c(block1, block2) return(res) }) blocks_liste_matSimilarity = unique(as.vector(blocks_liste_matSimilarityTemp1)) indice_blocks_liste_matSimilarityTemp = sapply(1:length(blocks_liste_matSimilarity), FUN = function(i){ res = which(res_block_splsda$names$blocks == blocks_liste_matSimilarity[i]) return(res) }) ind_Y = which(res_block_splsda$names$blocks == "Y") indice_blocks_liste_matSimilarity = indice_blocks_liste_matSimilarityTemp[indice_blocks_liste_matSimilarityTemp != ind_Y] boolean_pos_cor = TRUE if(length(indice_blocks_liste_matSimilarity) == 1) { }else{ for(i in 1:length(comp)) { comp_i = comp[i] for(j in 1:(length(indice_blocks_liste_matSimilarity) - 1)) { indice_blocks_liste_matSimilarity_j = indice_blocks_liste_matSimilarity[j] comp_indice_blocks_liste_matSimilarity_j = res_block_splsda$variates[[indice_blocks_liste_matSimilarity_j]][, comp_i] for(k in (j + 1):length(indice_blocks_liste_matSimilarity)) { indice_blocks_liste_matSimilarity_k = indice_blocks_liste_matSimilarity[k] comp_indice_blocks_liste_matSimilarity_k = res_block_splsda$variates[[indice_blocks_liste_matSimilarity_k]][, comp_i] cor = cor(comp_indice_blocks_liste_matSimilarity_j, comp_indice_blocks_liste_matSimilarity_k) boolean_pos_cor = boolean_pos_cor & cor > cutoff_comp } # Fin for(k in (j + 1):length(indice_blocks_vec_VarBlock)). } # Fin for(j in 1:(length(indice_blocks_vec_VarBlock)) - 1). } # Fin for(i in 1:length(comp)). } if(!boolean_pos_cor) { stop("For each pair of blocks, the ith component of the first block and the ith component of the second block have to be positively correlated in order to create a network.") }else{ liste_matSimilaritySelectTemp = lapply(1:length(liste_matSimilarity), FUN = function(j){ matSimilarity_j = liste_matSimilarity[[j]] indice_row_matSimilarity_j = which(rownames(matSimilarity_j)%in%vec_var == TRUE) indice_col_matSimilarity_j = which(colnames(matSimilarity_j)%in%vec_var == TRUE) if((length(indice_row_matSimilarity_j) != 0) &(length(indice_col_matSimilarity_j) != 0)) { res = matSimilarity_j[indice_row_matSimilarity_j, indice_col_matSimilarity_j, drop = FALSE] }else{ res = NA } return(res) }) names(liste_matSimilaritySelectTemp) = names(liste_matSimilarity) indice_NA_liste_matSimilaritySelectTemp = sapply(1:length(liste_matSimilaritySelectTemp), FUN = function(i){ liste_matSimilaritySelectTemp_i = liste_matSimilaritySelectTemp[[i]] if(is.matrix(liste_matSimilaritySelectTemp_i)) { res = FALSE }else{ if(is.na(liste_matSimilaritySelectTemp_i)) { res = TRUE }else{ res = FALSE } } return(res) }) liste_matSimilaritySelect = liste_matSimilaritySelectTemp[!indice_NA_liste_matSimilaritySelectTemp] w = c() node.X1 = c() node.X2 = c() vec_group = c() vec_nomsVar = c() for(i in 1:length(liste_matSimilaritySelect)) { noms_block1_block2 = names(liste_matSimilaritySelect)[i] matSimilaritySelect_i = liste_matSimilaritySelect[[i]] X1 = rownames(matSimilaritySelect_i) X2 = colnames(matSimilaritySelect_i) rep.X1 = rep(X1, each = length(X2)) rep.X2 = rep(X2, length(X1)) node.X1 = c(node.X1, rep.X1) node.X2 = c(node.X2, rep.X2) ch = strsplit(noms_block1_block2, split = "-")[[1]] nom_block1 = ch[1] nom_block2 = ch[2] vec_group = c(vec_group, c(rep(nom_block1, length(X1)), rep(nom_block2, length(X2)))) vec_nomsVar = c(vec_nomsVar, c(X1, X2)) w = c(w, as.vector(t(matSimilaritySelect_i))) } # Fin for(i in 1:length(liste_matSimilaritySelect)). dup = duplicated(vec_nomsVar) vec_nomsVar = vec_nomsVar[!dup] vec_group = vec_group[!dup] nodes = data.frame(name = vec_nomsVar, group = vec_group) # gR relations = data.frame(from = node.X1, to = node.X2, weight = w) # idx if(!is.null(vec_varBlock) & !is.null(vec_varBlockInteret2) & !is.null(vec_varRep)) { idx = sapply(1:dim(relations)[1], FUN = function(i){ node.X1_i = relations$from[i] node.X2_i = relations$to[i] if(node.X1_i%in%vec_varBlockInteret2 | node.X2_i%in%vec_varBlockInteret2) { res = TRUE }else if(node.X1_i%in%vec_varRep | node.X2_i%in%vec_varRep){ res = TRUE }else{ res = abs(w)[i] >= cutoff } return(res) }) }else if(!is.null(vec_varBlock) & is.null(vec_varBlockInteret2) & !is.null(vec_varRep)) { idx = sapply(1:dim(relations)[1], FUN = function(i){ node.X1_i = relations$from[i] node.X2_i = relations$to[i] if(node.X1_i%in%vec_varRep | node.X2_i%in%vec_varRep){ res = TRUE }else{ res = abs(w)[i] >= cutoff } return(res) }) }else if(is.null(vec_varBlockEtReponse) & !is.null(vec_varBlockInteret2) & !is.null(vec_varRep)) { idx = rep(TRUE, dim(relations)[1]) } relations = relations[idx, , drop = FALSE] gR = graph.data.frame(relations, directed = FALSE, vertices = nodes) # On supprime les noeuds qui n'ont pas d'arêtes. gR = delete.vertices(gR, which(degree(gR) == 0)) res = list(gR = gR) res$cutoff = cutoff return(res) } } }