d0L0

NAMESPACE

@@ -18,6 +18,7 @@ export(consensusSeqMIDAS2)
 export(consensusSeqMIDAS2Slow)
 export(doAdonis)
 export(doBoruta)
+export(doL0)
 export(drawEGGNOG)
 export(drawPATRIC)
 export(exportInteractive)

R/doL0.R

@@ -0,0 +1,58 @@
+#' doL0
+#' 
+#' feature selection of classification between category
+#' using snv frequency and gene matrix, based on L0Learn
+#' 
+#' @param stana stana object
+#' @param species candidate species ID
+#' @param cl named list of category
+#' @param target default to snps
+#' @param mat if target is not snps, provide preprocessed gene matrix
+#' otherwise the raw gene matrix is used.
+#' @param deleteZeroDepth delete zero depth SNV
+#' @param argList passed to L0Learn()
+#' @return L0Learn object
+#' @import ggplot2
+#' @export
+doL0 <- function(stana, species, cl=NULL, doFix=TRUE,
+    target="genes", mat=NULL, whichToCount="ec_description", argList=list(),
+    deleteZeroDepth=FALSE) {
+    if (length(argList)==0) {
+    	qqcat("Penalty is set to L0L2 by default")
+    	argList[["penalty"]] <- "L0L2"}
+    if (is.null(cl)) {
+        qqcat("Using grouping from the slot\n")
+        cl <- stana@cl
+    }
+    if (target!="snps" & is.null(mat)){
+        qqcat("If needed, please provide preprocessed matrix of genes to `mat`\n")
+        filtDf <- stana@genes[[species]]
+    } else if (target=="snps"){
+        filtDf <- stana@snps[[species]]
+        if (deleteZeroDepth) {
+            filtDf <- filtDf[rowSums(filtDf==-1)==0,]
+            qqcat("After filtering `-1`, position numbers: @{dim(filtDf)[1]}\n")
+        }
+    } else {
+        qqcat("Proceeding with provided matrix\n")
+        filtDf <- mat
+    }
+    qqcat("Feature number: @{dim(filtDf)[1]}\n")
+    transDf <- data.frame(t(filtDf), check.names=FALSE)
+    transDf <- transDf[intersect(row.names(transDf), cl |> unlist() |> unique()),]
+    gr <- NULL
+    for (cn in rownames(transDf)){
+        for (clm in seq_along(cl)){
+            if (cn %in% cl[[clm]]) {
+                gr <- c(gr, names(cl)[clm])
+            }
+        }
+    }
+
+
+    qqcat("Performing L0Learn\n")
+    argList[["x"]] <- transDf |> as.matrix()
+    argList[["y"]] <- as.factor(gr)
+    l0res <- do.call("L0Learn.cvfit", argList)
+    return(l0res)
+}