Add parameter sig_gene_threshold

Genes must be detected in at least `sig_gene_threshold`
proportion of cells to be considered when calculating
signature scores/backgrounds.

Author: deto

R/AnalysisFunctions.R

@@ -211,9 +211,11 @@ computeProjectionGenes <- function(object,
 #' on creating Signature objects.
 #' @param min_signature_genes Signature that match less than this number of genes in the
 #' supplied expression matrix are removed.
+#' @param sig_gene_threshold Proportion of cells that a gene must be detected in (nonzero)
+#' to be used in signature score calculations.
 #' @return the VISION object, with the @sigData slot updated
 #' @export
-addSignatures <- function(object, signatures, min_signature_genes=5) {
+addSignatures <- function(object, signatures, min_signature_genes=5, sig_gene_threshold=.01) {
 
     if (is.list(signatures)) {
         sigs <- lapply(signatures, function(sig){
@@ -237,7 +239,7 @@ addSignatures <- function(object, signatures, min_signature_genes=5) {
             Signature objects")
     }
 
-    sigs <- processSignatures(sigs, rownames(object@exprData), min_signature_genes)
+    sigs <- processSignatures(sigs, object@exprData, min_signature_genes, sig_gene_threshold)
 
     object@sigData <- c(object@sigData, sigs)
 

R/methods-Signature.R

@@ -4,11 +4,31 @@
 #' Drops signatures with less than `minSignatureGenes` matching genes
 #'
 #' @param sigData list of signature objects
-#' @param expressionGenes list of gene identifiers in the expression matrix
+#' @param exprData gene expression matrix
 #' @param minSignatureGenes minimum number of genes a signature must match
 #' in the expression matrix in order to be retained
+#' @param sig_gene_threshold Proportion of cells that a gene must be detected in (nonzero)
+#' to be used in signature score calculations.
+#' @importFrom Matrix rowSums
 #' @return processedSigData list of signature objects
-processSignatures <- function(sigData, expressionGenes, minSignatureGenes){
+processSignatures <- function(sigData, exprData, minSignatureGenes, sig_gene_threshold){
+    expressionGenes <- rownames(exprData)
+
+
+    cell_threshold <- sig_gene_threshold * ncol(exprData)
+    gene_detects <- rowSums(exprData > 0)
+    valid_genes <- gene_detects >= cell_threshold
+
+    message(
+        sprintf(
+            "\nUsing %i/%i genes detected in %.2f%% of cells for signature analysis.",
+            sum(valid_genes), nrow(exprData), sig_gene_threshold*100)
+        )
+    message(
+        "See the `sig_gene_threshold` input to change this behavior.\n"
+        )
+
+    expressionGenes <- rownames(exprData)[valid_genes]
     out <- lapply(sigData, function(sig){
         validGenes <- names(sig@sigDict) %in% expressionGenes
         sig@sigDict <- sig@sigDict[validGenes]

R/methods-Vision.R

@@ -20,6 +20,8 @@
 #' genes to use when computing projections.
 #' @param min_signature_genes Signature that match less than this number of genes in the
 #' supplied expression matrix are removed.
+#' @param sig_gene_threshold Proportion of cells that a gene must be detected in (nonzero)
+#' to be used in signature score calculations.
 #' @param threshold Threshold to apply when using the 'threshold' or 'fano' projection genes filter.
 #' If greater than 1, this specifies the number of cells in which a gene must be detected
 #' for it to be used when computing PCA. If less than 1, this instead specifies the proportion of cells needed
@@ -75,7 +77,9 @@ setMethod("Vision", signature(data = "matrixORSparse"),
             function(data, signatures=list(),
                     proteinData=NULL,
                     unnormalizedData = NULL, meta=NULL,
-                    projection_genes=c("fano"), min_signature_genes=5,
+                    projection_genes=c("fano"),
+                    min_signature_genes=5,
+                    sig_gene_threshold=.005,
                     threshold=.05, perm_wPCA=FALSE,
                     projection_methods = c("tSNE30"),
                     sig_norm_method = c("znorm_columns", "none", "znorm_rows",
@@ -101,7 +105,16 @@ setMethod("Vision", signature(data = "matrixORSparse"),
             .Object@params$micropooling <- list()
 
             rownames(data) <- toupper(rownames(data))
-            data <- data[ !duplicated(rownames(data)), , drop = FALSE]
+            toRemove <- rownames(data)[duplicated(rownames(data))]
+            if (length(toRemove) > 0){
+                message(sprintf(
+                        "\nRemoving %i genes with duplicate IDs (ignoring case): %s\n",
+                        length(toRemove) + length(unique(toRemove)),
+                        paste(unique(toRemove), collapse = ", ")
+                        )
+                    )
+                data <- data[ !(rownames(data) %in% toRemove), , drop = FALSE]
+            }
             .Object@exprData <- data
 
             if (!is.null(unnormalizedData)){
@@ -198,7 +211,7 @@ setMethod("Vision", signature(data = "matrixORSparse"),
                     Signature objects")
             }
 
-            .Object@sigData <- processSignatures(.Object@sigData, rownames(.Object@exprData), min_signature_genes)
+            .Object@sigData <- processSignatures(.Object@sigData, .Object@exprData, min_signature_genes, sig_gene_threshold)
 
             if (!is.null(meta)) {
                 if(is.matrix(meta)){

docs/reference/VISION-class.html

@@ -37,7 +37,7 @@ test_that("Naive Sig Scores computed correctly", {
                      sep = "\t", header = TRUE)
   expr <- data.matrix(expr)
   sigList <- readSignaturesInput(c("test_data/published_signatures/h.all.v5.2.symbols.gmt"))
-  sigList <- processSignatures(sigList, rownames(expr), 1)
+  sigList <- processSignatures(sigList, expr, 1, .000001)
 
   normData <- getNormalizedCopySparse(expr, "none")