@@ -43,6 +43,9 @@

 #' @param stratified_pam logical. If TRUE then maximum ASW is separately 

 #'   computed for each biological-cross-batch stratum (accepts NAs), and a 

 #'   weighted average silhouette width is returned as PAM_SIL. Default FALSE.

+#' @param stratified_rle logical. If TRUE then rle metrics are separately 

+#'   computed for each biological-cross-batch stratum (accepts NAs), and

+#'   weighted averages are returned for RLE_MED & RLE_IQR. Default FALSE.

 #'   

 #' @importFrom class knn

 #' @importFrom fpc pamk

@@ -61,8 +64,9 @@

 #'   correlation between expression pcs and negative control gene factors.} 

 #'   \item{EXP_WV_COR}{ Maximum squared spearman correlation between expression

 #'   pcs and positive control gene factors.} \item{RLE_MED}{ The mean squared 

-#'   median Relative Log Expression (RLE).} \item{RLE_IQR}{ The variance of the

-#'   inter-quartile range (IQR) of the RLE.} }

+#'   median Relative Log Expression (RLE) (see stratified_rle argument).}

+#'   \item{RLE_IQR}{ The variance of the inter-quartile range (IQR) of the RLE

+#'   (see stratified_rle argument).} }

 #'   

 #' @examples 

 #' 

@@ -83,7 +87,8 @@ score_matrix <- function(expr, eval_pcs = 3,

                          qc_factors = NULL,

                          uv_factors = NULL,

                          wv_factors = NULL,

-                         is_log=FALSE, stratified_pam = FALSE){

+                         is_log=FALSE, stratified_pam = FALSE,

+                         stratified_rle = FALSE){

   if(any(is.na(expr) | is.infinite(expr) | is.nan(expr))){

     stop("NA/Inf/NaN Expression Values.")

@@ -242,13 +247,36 @@ score_matrix <- function(expr, eval_pcs = 3,

   }

   ## ----- RLE Measures

-  rle <- expr - rowMedians(expr)

-  

-  # Non-Zero Median RLE

-  RLE_MED <- mean(colMedians(rle)^2) # Variance of the median

-  

-  # Variable IQR RLE

-  RLE_IQR <- var(colIQRs(rle)) # Variance of the IQR

+  if(stratified_rle){

+    

+    biobatch = as.factor(paste(bio,batch,sep = "_"))

+    RLE_MED <- RLE_IQR <- 0

+    

+    for (cond in levels(biobatch)) {

+      is_cond = which(biobatch == cond)

+      cond_w = length(is_cond)

+      rle <- expr[,is_cond] - rowMedians(expr[,is_cond])

+      

+      # Non-Zero Median RLE

+      RLE_MED <- RLE_MED + cond_w * mean(colMedians(rle) ^ 2) # Variance of the median

+      

+      # Variable IQR RLE

+      RLE_IQR <- RLE_IQR + cond_w * var(colIQRs(rle)) # Variance of the IQR

+      

+    }

+    

+    RLE_MED <- RLE_MED / length(biobatch)

+    RLE_IQR <- RLE_IQR / length(biobatch)

+    

+  } else{

+    rle <- expr - rowMedians(expr)

+    

+    # Non-Zero Median RLE

+    RLE_MED <- mean(colMedians(rle) ^ 2) # Variance of the median

+    

+    # Variable IQR RLE

+    RLE_IQR <- var(colIQRs(rle)) # Variance of the IQR

+  }

   scores = c(BIO_SIL, BATCH_SIL, PAM_SIL,

              EXP_QC_COR, EXP_UV_COR, EXP_WV_COR,

@@ -50,7 +50,9 @@

 #' @param stratified_pam logical. If TRUE then maximum ASW for PAM_SIL is

 #'   separately computed for each biological-cross-batch stratum (accepting

 #'   NAs), and a weighted average is returned as PAM_SIL.

-#'   

+#' @param stratified_rle logical. If TRUE then rle metrics are separately 

+#'   computed for each biological-cross-batch stratum (accepts NAs), and

+#'   weighted averages are returned for RLE_MED & RLE_IQR. Default FALSE.

 #' @param verbose logical. If TRUE some messagges are printed.

 #' @param run logical. If FALSE the normalization and evaluation are not run,

 #'   but normalization parameters are returned in the output object for 

@@ -169,6 +171,7 @@ setMethod(

                         eval_proj = NULL,

                         eval_proj_args = NULL, eval_kclust=2:10,

                         verbose=FALSE, stratified_pam = FALSE,

+                        stratified_rle = FALSE,

                         return_norm = c("no", "in_memory", "hdf5"),

                         hdf5file,

                         bpparam=BiocParallel::bpparam()) {

@@ -329,6 +332,12 @@ setMethod(

         stop("'eval_kclust' must be less than the number of samples.")

       }

+      if(stratified_rle) {

+        if(is.null(bio) & is.null(batch)){

+          stop("For stratified_rle, bio and/or batch must be specified")

+        }

+      }

+      

       if(!is.null(eval_kclust) & stratified_pam) {

         if(is.null(bio) & is.null(batch)){

           stop("For stratified_pam, bio and/or batch must be specified")

@@ -15,8 +15,8 @@ scone(x, ...)

   adjust_batch = c("no", "yes", "force"), run = TRUE, evaluate = TRUE,

   eval_pcs = 3, eval_proj = NULL, eval_proj_args = NULL,

   eval_kclust = 2:10, verbose = FALSE, stratified_pam = FALSE,

-  return_norm = c("no", "in_memory", "hdf5"), hdf5file,

-  bpparam = BiocParallel::bpparam())

+  stratified_rle = FALSE, return_norm = c("no", "in_memory", "hdf5"),

+  hdf5file, bpparam = BiocParallel::bpparam())

 }

 \arguments{

 \item{x}{a \code{\link{SconeExperiment}} object.}

@@ -77,6 +77,10 @@ If NULL, tightness will be returned NA.}

 separately computed for each biological-cross-batch stratum (accepting

 NAs), and a weighted average is returned as PAM_SIL.}

+\item{stratified_rle}{logical. If TRUE then rle metrics are separately 

+computed for each biological-cross-batch stratum (accepts NAs), and

+weighted averages are returned for RLE_MED & RLE_IQR. Default FALSE.}

+

 \item{return_norm}{character. If "no" the normalized values will not be

 returned with the output object. This will create a much smaller object

 and may be useful for large datasets and/or when many combinations are

@@ -7,7 +7,7 @@

 score_matrix(expr, eval_pcs = 3, eval_proj = NULL, eval_proj_args = NULL,

   eval_kclust = NULL, bio = NULL, batch = NULL, qc_factors = NULL,

   uv_factors = NULL, wv_factors = NULL, is_log = FALSE,

-  stratified_pam = FALSE)

+  stratified_pam = FALSE, stratified_rle = FALSE)

 }

 \arguments{

 \item{expr}{matrix. The expression data matrix (genes in rows, cells in 

@@ -50,6 +50,10 @@ transformation will not be carried out prior to projection. Default FALSE.}

 \item{stratified_pam}{logical. If TRUE then maximum ASW is separately 

 computed for each biological-cross-batch stratum (accepts NAs), and a 

 weighted average silhouette width is returned as PAM_SIL. Default FALSE.}

+

+\item{stratified_rle}{logical. If TRUE then rle metrics are separately 

+computed for each biological-cross-batch stratum (accepts NAs), and

+weighted averages are returned for RLE_MED & RLE_IQR. Default FALSE.}

 }

 \value{

 A list with the following metrics: \itemize{ \item{BIO_SIL}{ Average

@@ -61,8 +65,9 @@ A list with the following metrics: \itemize{ \item{BIO_SIL}{ Average

   correlation between expression pcs and negative control gene factors.} 

   \item{EXP_WV_COR}{ Maximum squared spearman correlation between expression

   pcs and positive control gene factors.} \item{RLE_MED}{ The mean squared 

-  median Relative Log Expression (RLE).} \item{RLE_IQR}{ The variance of the

-  inter-quartile range (IQR) of the RLE.} }

+  median Relative Log Expression (RLE) (see stratified_rle argument).}

+  \item{RLE_IQR}{ The variance of the inter-quartile range (IQR) of the RLE

+  (see stratified_rle argument).} }

 }

 \description{

 This function evaluates a (normalized) expression matrix using SCONE