account for missing injections in sbc (#30)

* dont supress warnings for knn
* signal correction returns qc fit in processing history
* batch correction uses correct order for samples

Author: grlloyd

R/mv_imputation.R

@@ -131,7 +131,7 @@ mv_imputation <- function(df, method, k=10, rowmax=0.5, colmax=0.5,
     remove these columns using the sample filter tool")
     }
     if (tolower(method) == "knn") {
-        obj <- suppressWarnings(impute.knn(assay(df), k=k, 
+        obj <- (impute.knn(assay(df), k=k, 
             rowmax=rowmax, colmax=colmax, maxp=maxp))
         assay(df) <- obj$data
     } else if (tolower(method) == "rf") {

R/sbc_main.R

@@ -66,6 +66,10 @@ QCRSC <- function(df, order, batch, classes, spar = 0, log = TRUE,
         log = log, spar = spar, batch = batch, minQC = minQC)
     QC_fit <- do.call(rbind, QC_fit)
 
+    meta_data <- metadata(df)
+    meta_data$processing_history$QCRSC <- return_function_args()
+    meta_data$processing_history$QCRSC$QC_fit=QC_fit
+    
     # Median value for each fature, and divide it by predicted value
     mpa <- matrixStats::rowMedians(assay(df), na.rm=TRUE)
     QC_fit <- QC_fit/mpa
@@ -74,8 +78,7 @@ QCRSC <- function(df, order, batch, classes, spar = 0, log = TRUE,
     assay(df) <- assay(df)/QC_fit
     assay(df)[assay(df) <= 0] <- NA
 
-    meta_data <- metadata(df)
-    meta_data$processing_history$QCRSC <- return_function_args()
+    
     metadata(df) <- meta_data
     df <- return_original_data_structure(df)
 

R/sbc_methods.R

@@ -79,39 +79,43 @@ sbcWrapper <- function(id, qcData, order, qcBatch, qcOrder, log=log, spar=spar,
     batch=batch, minQC) {
 
     out <- tryCatch ({
-    # Measurment order can be non consecutive numbers as well
-    maxOrder <- length(order)
-    newOrder <- seq_len(maxOrder)
+
+    # qc data for feature
     subData <- qcData[id, ]
+    
+    # unique batches
     nbatch <- unique(qcBatch)
 
-    outl <- matrix(nrow=maxOrder, ncol=length(nbatch))
+    # preallocate output
+    outl <- numeric(length(order))
 
+    # for each batch
     for (nb in seq_len(length(nbatch))) {
+        # get the injection order for the QCs
         x <- qcOrder[qcBatch == nbatch[nb]]
+        # get the response for the QCs
         y <- subData[qcBatch == nbatch[nb]]
+        
+        # remove any index with NA in the response
         NAhits <- which(is.na(y))
 
         if (length(NAhits) > 0) {
             x <- x[-c(NAhits)]
             y <- y[-c(NAhits)]
         }
 
+        # if we have enough QCs values
         if (length(y) >= minQC) {
-            outl[,nb] <- splineSmoother(x=x, y=y, newX=newOrder, log=log,
+            # fit spline to QCs and get predictions for all samples in batch
+            outl[batch==nbatch[nb]] <- splineSmoother(x=x, y=y, newX=order[batch==nbatch[nb]], log=log,
             a=1, spar=spar)
         } else {
-            outl[,nb] <- rep(NA, maxOrder)
+            # otherwise replace with NA
+            outl[batch==nbatch[nb]] = NA
         }
     }
 
-    outp <- rep(NA, nrow(outl))
-    for (nb in seq_len(length(nbatch))) {
-        range <- c(batch == nbatch[nb])
-        outp[range] <- outl[range, nb]
-    }
-
-    outp
+    outp=outl
     },
 
     error = function(e){