Commit 2c3694af authored by Dimitrios Kyriakis's avatar Dimitrios Kyriakis
Browse files

readme

parent 96a9598c
...@@ -64,7 +64,6 @@ set.seed(123) ...@@ -64,7 +64,6 @@ set.seed(123)
<details><summary>Code</summary> <details><summary>Code</summary>
<p> <p>
```{r setup} ```{r setup}
# ================================ SETTING UP ======================================== #
tool="seurat" tool="seurat"
project ="Michi_Data" project ="Michi_Data"
dataset <- project dataset <- project
...@@ -82,7 +81,6 @@ color_cond <- c( "magenta4", "#007A87",brewer.pal(6,"Dark2")[-1],"#FF5A5F","blac ...@@ -82,7 +81,6 @@ color_cond <- c( "magenta4", "#007A87",brewer.pal(6,"Dark2")[-1],"#FF5A5F","blac
color_clust <- c(brewer.pal(12,"Paired")[-11],"black","gray","magenta4","seagreen4",brewer.pal(9,"Set1")[-6],brewer.pal(8,"Dark2")) color_clust <- c(brewer.pal(12,"Paired")[-11],"black","gray","magenta4","seagreen4",brewer.pal(9,"Set1")[-6],brewer.pal(8,"Dark2"))
color_cells <- c(brewer.pal(9,"Set1")[-6],"goldenrod4","darkblue","seagreen4") color_cells <- c(brewer.pal(9,"Set1")[-6],"goldenrod4","darkblue","seagreen4")
color_list <- list(condition=color_cond,Cluster=color_clust,Cell_Type=color_cells,State=color_clust) color_list <- list(condition=color_cond,Cluster=color_clust,Cell_Type=color_cells,State=color_clust)
# ========= Parameters
imputation = FALSE imputation = FALSE
remove_mt=FALSE remove_mt=FALSE
remove_ribsomal=FALSE remove_ribsomal=FALSE
...@@ -106,7 +104,6 @@ Additional to this filtering, we defined cells as low-quality, based on three cr ...@@ -106,7 +104,6 @@ Additional to this filtering, we defined cells as low-quality, based on three cr
<details><summary>Code</summary> <details><summary>Code</summary>
<p> <p>
```{r readfiles} ```{r readfiles}
# ======== Perform an integrated analysis ====
NewDir <- paste0(Sys.Date(),"_",tool,"_elbow_",elbow,"_Mito-",remove_mt,"_Ribo-",remove_ribsomal,"_SCT-",SCT,"_criteria_pass-",criteria_pass) NewDir <- paste0(Sys.Date(),"_",tool,"_elbow_",elbow,"_Mito-",remove_mt,"_Ribo-",remove_ribsomal,"_SCT-",SCT,"_criteria_pass-",criteria_pass)
dir.create(NewDir) dir.create(NewDir)
setwd(NewDir) setwd(NewDir)
...@@ -141,7 +138,6 @@ The integration of the filtered matrices of the different datasets was performed ...@@ -141,7 +138,6 @@ The integration of the filtered matrices of the different datasets was performed
```{r remapping} ```{r remapping}
dir.create("Aligned_Cond_RegPhase") dir.create("Aligned_Cond_RegPhase")
setwd("Aligned_Cond_RegPhase") setwd("Aligned_Cond_RegPhase")
# ================================== ALLIGN CONDITIONS =========================================
DefaultAssay(Combined) <- "RNA" DefaultAssay(Combined) <- "RNA"
Combined$condition <- factor(as.factor(Combined$condition), levels = c("Control_IPSCs", "Control_D06" ,"Control_D10", "Control_D15", "Control_D21", Combined$condition <- factor(as.factor(Combined$condition), levels = c("Control_IPSCs", "Control_D06" ,"Control_D10", "Control_D15", "Control_D21",
"PINK1_IPSCs","PINK1_D06", "PINK1_D15", "PINK1_D21")) "PINK1_IPSCs","PINK1_D06", "PINK1_D15", "PINK1_D21"))
...@@ -152,7 +148,6 @@ pink.list <-SplitObject(Combined,split.by = "Treatment") ...@@ -152,7 +148,6 @@ pink.list <-SplitObject(Combined,split.by = "Treatment")
for (i in 1:length(pink.list)) { for (i in 1:length(pink.list)) {
pink.list[[i]] <- SCTransform(pink.list[[i]], verbose = FALSE,vars.to.regress=c("G2M.Score","S.Score")) pink.list[[i]] <- SCTransform(pink.list[[i]], verbose = FALSE,vars.to.regress=c("G2M.Score","S.Score"))
} }
# doi: https://doi.org/10.1101/576827
int.features <- SelectIntegrationFeatures(object.list = pink.list, nfeatures = 3000) int.features <- SelectIntegrationFeatures(object.list = pink.list, nfeatures = 3000)
pink.list <- PrepSCTIntegration(object.list = pink.list, anchor.features = int.features, pink.list <- PrepSCTIntegration(object.list = pink.list, anchor.features = int.features,
verbose = FALSE) verbose = FALSE)
...@@ -178,7 +173,6 @@ The clustering of data was performed using Louvain clustering. The resolution of ...@@ -178,7 +173,6 @@ The clustering of data was performed using Louvain clustering. The resolution of
<details><summary>Code</summary> <details><summary>Code</summary>
<p> <p>
```{r Clustering} ```{r Clustering}
# ================================== Clustering =========================================
dir.create("Clusters") dir.create("Clusters")
setwd("Clusters") setwd("Clusters")
Combined <- ICSWrapper::reduce_dim(Combined,project=project,assay = "SCT")$Combined#,resolution=c(0.1))$Combined Combined <- ICSWrapper::reduce_dim(Combined,project=project,assay = "SCT")$Combined#,resolution=c(0.1))$Combined
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment