@@ -440,6 +440,7 @@ parent <- function(classifier) {

     # set new features

     new_features <- labels(value$terms)

+    new_features <- gsub('_', '-', new_features) # convert underscore to hyphen if exists

     features(classifier) <- new_features

   } else {

     stop("Can only assign new classifier for a cell type that has no parent.

@@ -155,6 +155,9 @@ setMethod("train_classifier", c("train_obj" = "Seurat"),

   # transform list to factor

   train_tag <- factor(train_tag, levels = c('yes', 'no'))

+  # convert hyphen (-) by underscore (_)

+  colnames(mat) <- gsub('-', '_', colnames(mat))

+  

   # train

   clf <- train_func(mat, train_tag)

@@ -162,7 +165,9 @@ setMethod("train_classifier", c("train_obj" = "Seurat"),

   clf$resampledCM <- NULL 

   p_thres <- 0.5

-  object <- scClassifR(cell_type, clf, labels(clf$terms), p_thres, 

+  features <- labels(clf$terms)

+  features <- gsub('_', '-', features) # convert back underscore to hyphen

+  object <- scClassifR(cell_type, clf, features, p_thres, 

                              NA_character_)

   # only assign parent if pretrained model for parent cell type is avai

@@ -256,15 +261,19 @@ setMethod("train_classifier", c("train_obj" = "SingleCellExperiment"),

   # transform list to factor

   train_tag <- factor(train_tag, levels = c('yes', 'no'))

+  # convert hyphen (-) by underscore (_)

+  colnames(mat) <- gsub('-', '_', colnames(mat))

+  

   # train

   clf <- train_func(mat, train_tag)

-  

   # remove this info to reduce memory

   clf$resampledCM <- NULL 

   p_thres <- 0.5

-  object <- scClassifR(cell_type, clf, labels(clf$terms), p_thres, 

+  features <- labels(clf$terms)

+  features <- gsub('_', '-', features) # convert back underscore to hyphen

+  object <- scClassifR(cell_type, clf, features, p_thres, 

                              NA_character_)

   # only assign parent if pretrained model for parent cell type is avai

@@ -497,6 +497,9 @@ setMethod("process_parent_clf", c("obj" = "Seurat"),

         filtered_mat <- transform_to_zscore(filtered_mat)

       }

+      # to avoid problem triggered by '-' in gene names

+      colnames(filtered_mat) <- gsub('-', '_', colnames(filtered_mat))

+      

       # predict

       pred = stats::predict(clf(parent.clf), filtered_mat, type = "prob") %>% 

            dplyr::mutate('class' = apply(., 1, 

@@ -581,6 +584,9 @@ setMethod("process_parent_clf", c("obj" = "SingleCellExperiment"),

         filtered_mat <- transform_to_zscore(filtered_mat)

       }

+      # to avoid problem triggered by '-' in gene names

+      colnames(filtered_mat) <- gsub('-', '_', colnames(filtered_mat))

+      

       # predict

       pred = stats::predict(clf(parent.clf), filtered_mat, type = "prob") %>% 

         dplyr::mutate('class' = apply(., 1, 

@@ -626,6 +632,9 @@ make_prediction <- function(mat, classifier, pred_cells,

   . <- NULL

   cells <- names(pred_cells)

+  # to avoid problem triggered by '-' in gene names

+  colnames(mat) <- gsub('-', '_', colnames(mat))

+  

   # predict

   pred = stats::predict(clf(classifier), mat, type = "prob") %>%

     dplyr::mutate('class' = apply(., 1, function(x)

@@ -773,6 +782,10 @@ verify_parent <- function(mat, classifier, meta.data) {

 test_performance <- function(mat, classifier, tag) {

   overall.roc <- . <- NULL

+  # to avoid problem triggered by '-' in gene names

+  colnames(mat) <- gsub('-', '_', colnames(mat))

+  #labels(clf(classifier)$terms) <- gsub('-', '_', labels(clf(classifier)$terms))

+    

   tag <- unlist(lapply(tag, function(x) if (x == 'yes') {1} else {0}))

   iter <- unique(sort(c(p_thres(classifier), seq(0.1, 0.9, by = 0.1))))