Added variance filtering to custom/matrixfilter (#5358)

* Added variant filtering to matrix filter

* Corrected indentation

* Simplified function in single conditional block

* fixed line indentation

* Fixed whitespace

* Fixed logic

* line wrapping change

* added whitespace around comments

* [automated] Fix linting with Prettier

* remove trailing whitespace

* Update matrixfilter.R

* Remove whitespace

---------

Co-authored-by: Jonathan Manning <jonathan.manning@seqera.io>
Co-authored-by: nf-core-bot <core@nf-co.re>
Co-authored-by: Jonathan Manning <pininforthefjords@gmail.com>
Co-authored-by: Simon Pearce <24893913+SPPearce@users.noreply.github.com>

modules/nf-core/custom/matrixfilter/meta.yml

@@ -72,6 +72,12 @@ input:
         Minimum observations that must have a numeric value (not be NA) to retain
         the row/ feature (e.g. gene). Supplied via task.ext.args as
         --minimum_samples_not_na. Overrides minimum_proportion_not_na
+  - most_variant_features:
+      type: integer
+      description: |
+        Variance filter for the number of  row/ feature (e.g. gene) observations returned.
+        Supplied via task.ext.args as --most_variant_features
+
 output:
   - versions:
       type: file

modules/nf-core/custom/matrixfilter/templates/matrixfilter.R

@@ -69,6 +69,21 @@ read_delim_flexible <- function(file, header = TRUE, row.names = NULL, nrows = -
     )
 }
 
+#' Identify rows that are among the top n most variant
+#'
+#' @param matrix_data Matrix object
+#'
+#' @return output Boolean vector
+
+most_variant_test <- function(matrix_data) {
+
+    # Determine the indices of the top variant rows based on variance
+    top_indices <- order(-apply(matrix_data, 1, var, na.rm = TRUE))[1:opt\$most_variant_features]
+
+    # Return a boolean vector indicating if each row is among the top variant ones
+    1:nrow(matrix_data) %in% top_indices
+}
+
 # Set up default options
 
 opt <- list(
@@ -80,7 +95,8 @@ opt <- list(
     minimum_proportion = 0,
     grouping_variable = NULL,
     minimum_proportion_not_na = 0.5,
-    minimum_samples_not_na = NULL
+    minimum_samples_not_na = NULL,
+    most_variant_features = NULL
 )
 opt_types <- lapply(opt, class)
 
@@ -176,6 +192,14 @@ boolean_matrix <- t(apply(abundance_matrix, 1, function(row) {
     sapply(tests, function(f) f(row))
 }))
 
+# Apply the 'most_variant_test' function to identify the most variant rows and add
+# the result to the boolean matrix
+
+if (! is.null(opt\$most_variant_features)) {
+    most_variant_vectors <- most_variant_test(abundance_matrix)
+    boolean_matrix <- cbind(boolean_matrix, most_variant_vectors)
+}
+
 # We will retain features passing all tests
 
 keep <- apply(boolean_matrix, 1, all)

tests/modules/nf-core/custom/matrixfilter/main.nf

@@ -71,3 +71,15 @@ workflow test_custom_matrixfilter_na_samples {
         ch_samplesheet
     )
 }
+
+workflow test_custom_matrixfilter_var {
+
+    expression_matrix_file = file(params.test_data['mus_musculus']['genome']['rnaseq_matrix'], checkIfExists: true)
+
+    ch_samples_matrix = [ [ "id":"SRP254919" ], expression_matrix_file ]
+
+    CUSTOM_MATRIXFILTER(
+        ch_samples_matrix,
+        empty_samplesheet
+    )
+}

tests/modules/nf-core/custom/matrixfilter/nextflow.config

@@ -26,4 +26,8 @@ process {
         ext.args = { "--sample_id_col Experiment --minimum_abundance 28 --minimum_samples_not_na 1" }
         ext.prefix = { "${meta.id}_test" }
     }
+     withName: 'test_custom_matrixfilter_var:CUSTOM_MATRIXFILTER' {
+        ext.args = { "--most_variant_features 500" }
+        ext.prefix = { "${meta.id}_test" }
+    }
 }