Change: Use caret for one-hot encoding sample_metadata in mldm

micone/pipelines/envs/micone-mldm/env.yml

@@ -6,5 +6,6 @@ channels:
 dependencies:
     - python=3.8
     - r-base
+    - r-caret
     - openjdk
     - pip>=20.2.4

micone/pipelines/templates/network_inference/direct/mldm.R

@@ -1,23 +1,27 @@
 #!/usr/bin/env Rscript
 
 library(mLDM)
+library(caret)
 
 otu_file <- "${otu_file}"
 sample_metadata <- "${sample_metadata}"
 Z_mean <- ${Z_mean}
 max_iteration <- ${max_iteration}
 corr_file <- "${meta.id}_corr.tsv"
 
-read_data <- function(datafile) {
-    data <- read.table(datafile, header=TRUE, comment.char="", sep="\\t")
-    data.rownames <- data[, 1]
-    data <- data.matrix(data[, 2:ncol(data)])
-    rownames(data) <- data.rownames
-    return(data)
+read_data <- function(tablefile) {
+    table <- read.table(tablefile, header=TRUE, comment.char="", sep="\\t")
+    table.rownames <- table[, 1]
+    table <- table[, 2:ncol(table)]
+    dmy <- dummyVars(" ~ .", data=table)
+    table.frame <- data.frame(predict(dmy, newdata=table))
+    table.matrix <- data.matrix(table.frame)
+    rownames(table.matrix) <- table.rownames
+    return(table.matrix)
 }
 
 otu <- read_data(otu_file)
-sample.md <- read_data(sample_metadata) # Must be purely numeric metadata
+sample.md <- read_data(sample_metadata) # Must be purely numeric or categorical metadata
 
 mldmNetwork <- mLDM(X=t(otu), M=sample.md, Z_mean=Z_mean, max_iteration=max_iteration, verbose=TRUE)