Add SingleR and CellAssign metadata

add-celltypes.nf

@@ -11,8 +11,8 @@ if (!file(params.run_metafile).exists()) {
   param_error = true
 }
 
-if (!file(params.celltype_refs_metafile).exists()) {
-  log.error("The 'celltype_refs_metafile' file '${params.celltype_refs_metafile}' can not be found.")
+if (!file(params.celltype_project_metafile).exists()) {
+  log.error("The 'celltype_project_metafile' file '${params.celltype_project_metafile}' can not be found.")
   param_error = true
 }
 

bin/save_singler_refs.R

@@ -0,0 +1,33 @@
+#!/usr/bin/env Rscript
+
+# this script is used to download celldex references and save them as RDS files
+
+library(optparse)
+library(celldex)
+
+option_list <- list(
+  make_option(
+    opt_str = c("--ref_name"),
+    type = "character",
+    help = "name associated with celldex reference, e.g., HumanPrimaryCellAtlasData"
+  ),
+  make_option(
+    opt_str = c("--ref_file"),
+    type = "character",
+    help = "path to save reference file"
+  )
+)
+
+# Parse options
+opt <- parse_args(OptionParser(option_list = option_list))
+
+# check that provided ref name is in celldex package
+if(! opt$ref_name %in% ls("package:celldex")){
+  stop(glue::glue("Provided `ref_name` `{opt$ref_name}` does not match a celldex dataset."))
+} 
+
+# get a reference library from celldex:
+ref <- do.call(`::`, args = list("celldex", opt$ref_name))(ensembl = TRUE)
+
+# export ref data 
+readr::write_rds(ref, opt$ref_file, compress = "gz")

build-celltype-ref.nf

@@ -3,42 +3,88 @@ nextflow.enable.dsl=2
 
 params.t2g_3col_path = "s3://scpca-references/homo_sapiens/ensembl-104/annotation/Homo_sapiens.GRCh38.104.spliced_intron.tx2gene_3col.tsv"
 
+process save_singler_refs {
+  container params.SCPCATOOLS_CONTAINER
+  publishDir "${params.singler_references_dir}"
+  label 'mem_8'
+  input:
+    tuple val(ref_name), val(ref_source)
+  output:
+    tuple val(ref_name), path(ref_file)
+  script:
+    ref_file = "${ref_source}-${ref_name}.rds"
+    """
+    save_singler_refs.R \
+     --ref_name ${ref_name} \
+     --ref_file ${ref_file}
+    """
+  stub:
+    ref_file = "${ref_source}-${ref_name}.rds"
+    """
+    touch ${ref_file}
+    """
+
+}
+
 process train_singler_models {
   container params.SCPCATOOLS_CONTAINER
-  publishDir "${params.celltype_model_dir}"
+  publishDir "${params.singler_models_dir}"
   label 'cpus_4'
   label 'mem_16'
   input:
-    tuple path(celltype_ref), val(ref_name)
+    tuple val(ref_name), path(ref_file)
     path tx2gene
   output:
     path celltype_model
   script:
     celltype_model = "${ref_name}_model.rds"
     """
     train_SingleR.R \
-      --ref_file ${celltype_ref} \
+      --ref_file ${ref_file} \
       --output_file ${celltype_model} \
       --fry_tx2gene ${tx2gene} \
       --label_name ${params.label_name} \
       --seed ${params.seed} \
       --threads ${task.cpus}
     """
+  stub:
+    celltype_model = "${ref_name}_model.rds"
+    """
+    touch ${celltype_model}
+    """
 }
 
 workflow build_celltype_ref {
 
   // create channel of cell type ref files and names
-  celltype_refs_ch = Channel.fromPath(params.celltype_refs_metafile)
+  celltype_refs_ch = Channel.fromPath(params.celltype_ref_metadata)
     .splitCsv(header: true, sep: '\t')
+    .branch{
+          singler: it.celltype_method == "SingleR"
+          cellassign: it.celltype_method == "CellAssign"
+       }
+
+  // singler refs to download and train
+  singler_refs_ch = celltype_refs_ch.singler
     .map{[
-      celltype_ref_file = "${params.celltype_ref_dir}/${it.celltype_singler_file}",
-      ref_name = it.celltype_ref_name
+      ref_name: it.celltype_ref_name,
+      ref_source: it.celltype_ref_source
       ]}
-    .unique() // remove any duplicates
+
+  // download and save reference files
+  save_singler_refs(singler_refs_ch)
 
   // train cell type references using SingleR
-  train_singler_models(celltype_refs_ch, params.t2g_3col_path)
+  train_singler_models(save_singler_refs.out, params.t2g_3col_path)
+
+  // cellassign refs
+  cellassign_refs_ch = celltype_refs_ch.cellassign
+    // create a channel with ref_name, source, organs
+    .map{[
+      ref_name: it.celltype_ref_name,
+      ref_source: it.celltype_ref_source,
+      organs: it.organs
+    ]}
 }
 
 workflow {

config/profile_ccdl.config

@@ -13,7 +13,7 @@ params{
   run_ids = "SCPCR000001,SCPCS000101"
 
   // cell type references and SingleR label of interest
-  celltype_refs_metafile = "s3://ccdl-scpca-data/sample_info/celltype_annotation/scpca-project-celltype-metadata.tsv"
+  celltype_project_metafile = "s3://ccdl-scpca-data/sample_info/celltype_annotation/scpca-project-celltype-metadata.tsv"
   label_name = "label.ont"
 
 

config/reference_paths.config

@@ -6,6 +6,13 @@ ref_rootdir  = 's3://scpca-references'
 // barcode files
 barcode_dir  = "${params.ref_rootdir}/barcodes/10X"
 
-// cell type references
-celltype_ref_dir = "${params.ref_rootdir}/celltype/references" // original (generally celldex) files themselves
-celltype_model_dir = "${params.ref_rootdir}/celltype/singler_models" // models build from original (celldex) files
+// cell type references directories
+celltype_ref_dir = "${params.ref_rootdir}/celltype"
+// output from save_singler_refs() process
+singler_references_dir = "${params.celltype_ref_dir}/singler_references"
+// output from train_singler_models() process
+singler_models_dir = "${params.celltype_ref_dir}/singler_models"
+
+// cell type metadata
+celltype_ref_metadata = "${projectDir}/references/celltype-reference-metadata.tsv"
+panglao_marker_genes_file = "${projectDir}/references/PanglaoDB_markers_27_Mar_2020.tsv"

modules/classify-celltypes.nf

@@ -30,7 +30,7 @@ workflow annotate_celltypes {
     take: processed_sce_channel
     main:
       // channel with celltype model and project ids
-      celltype_ch = Channel.fromPath(params.celltype_refs_metafile)
+      celltype_ch = Channel.fromPath(params.celltype_project_metafile)
         .splitCsv(header: true, sep: '\t')
         .map{[
           project_id = it.scpca_project_id,