Merge remote-tracking branch 'origin/master'

chipseq_workflow/chipseq_utils.sh

@@ -16,7 +16,7 @@ export PYTHONPATH=/home/brandl/bin/macs2/lib/python2.7/site-packages:$PYTHONPATH
 
 cs_bowtie_qc(){
 
-rendr_snippet  "bowtie2_qc" <<EOF
+rendr_snippet  "bowtie2_qc" <<"EOF"
 devtools::source_url("https://raw.githubusercontent.com/holgerbrandl/datautils/v1.8/R/core_commons.R")
 devtools::source_url("https://raw.githubusercontent.com/holgerbrandl/datautils/v1.8/R/ggplot_commons.R")
 

dge_workflow/limma/dge_limma.R

@@ -45,8 +45,7 @@ gene_info_file = opts$gene_info
 assert(is.null(gene_info_file) || file.exists(gene_info_file), "invalid gene_info_file")
 
 designFormula = opts$design
-assert(str_detect(designFormula, ".*condition$"))
-
+assert(str_detect(designFormula, "^condition.*")) ## make sure that the condition comes before all batch factors
 
 results_prefix = if (str_length(opts$out) > 0) opts$out else "" # used by add_prefix
 
@@ -136,7 +135,21 @@ group_labels = data_frame(replicate = colnames(expMatrix)) %>%
 names(group_labels) = colnames(expMatrix)
 makePcaPlot(t(expMatrix), color_by = group_labels, title = "PCA of quantifiable proteins in all conditions")
 
+#' Also do a scatter plot matrix of the PCs
+mydata.pca = prcomp(t(expMatrix), retx = TRUE, center = TRUE, scale. = FALSE)
+
 
+# screeplot(mydata.pca)
+# devtools::install_github("vqv/ggbiplot")
+# require(ggbiplot)
+ggbiplot::ggscreeplot(mydata.pca) + geom_col()
+
+# load_pack(GGally)
+pcs = mydata.pca$x %>% as_df %>% rownames_to_column("sample")
+pcs %>% GGally::ggpairs(columns=2:6, mapping=ggplot2::aes(color=sample), upper="blank", legend=c(3,3)) + theme(legend.position = "bottom")
+
+
+#' Also analyze spearman correlation
 correlation = cor(expMatrix, method = "spearman")
 library(lattice)
 levelplot(correlation, scales = list(x = list(rot = 90)), pretty = TRUE, main = "Spearman correlation between conditions after Normalization", xlab = "Conditions", ylab = "Conditions")
@@ -180,11 +193,15 @@ orderMatcheExpDesign = data_frame(replicate = colnames(expMatrix)) %>%
     right_join(expDesign, by = "replicate") %>%
     arrange(col_index)
 
-## build design matrix
-#A key strength of limma’s linear modelling approach, is the ability accommodate arbitrary experimental complexity. Simple designs, such as the one in this workflow, with cell type and batch, through to more complicated factorial designs and models with interaction terms can be handled relatively easily
+#' Build design matrix
+#' > A key strength of limma’s linear modelling approach, is the ability accommodate arbitrary experimental complexity. Simple designs, such as the one in this workflow, with cell type and batch, through to more complicated factorial designs and models with interaction terms can be handled relatively easily
+#'
+#' Make sure that non of the batch-factors  is confounded with treatment (condition). See https://support.bioconductor.org/p/39385/ for a discussion
+#' References
+#' * https://f1000research.com/articles/5-1408/v1
 
-# design <- orderMatcheExpDesign %$% model.matrix(~ 0 +  condition)
-design <- orderMatcheExpDesign %$% model.matrix(formula(as.formula(paste("~0+", designFormula))))
+# design <- orderMatcheExpDesign %$% model.matrix(~ 0 +  condition + prep_day)
+design = orderMatcheExpDesign %$% model.matrix(formula(as.formula(paste("~0+", designFormula))))
 rownames(design) <- orderMatcheExpDesign$replicate
 
 #design <- model.matrix(~0+group+lane)
@@ -319,7 +336,8 @@ deResults %<>% left_join(sampleMeans)
 
 #' apply the hit criterion
 
-deResults %>% ggplot(aes(adj_p_val)) + geom_histogram()
+deResults %>% ggplot(aes(p_value)) + geom_histogram(binwidth=.01)
+deResults %>% ggplot(aes(adj_p_val)) + geom_histogram(binwidth=.01)
 
 # report hit criterion
 #+ results='asis'