library(reticulate)
reticulate::install_miniconda()
reticulate::use_condaenv()
py_install("python-igraph")
py_install("numpy")
py_install("leidenalg", forge = TRUE)
library(DIscBIO)
library(leiden)
library(M3Drop)


load("SC.RData")           # Loading the "SC" object 
filteredDataset<-SC@fdata
xprs <- as.matrix(filteredDataset)
xprs[xprs == min(xprs)] <- 0
top.feats <- M3Drop::M3DropFeatureSelection(expr_mat = xprs, mt_method = "fdr",suppress.plot = FALSE, mt_threshold = 0.5) ## Fits a Michaelis-Menten function to the dropout-rate.
top.feats <- head(top.feats$Gene, 500)
top.xprs <- xprs[top.feats, ]
Adj <- stats::cor(top.xprs)
QNT <- quantile(Adj, probs = c(0.5))
bin.Adj <- (Adj > QNT) + 0

lei.clusts <- leiden(bin.Adj, resolution_parameter = 0.5)   # Leiden algorithm
names(lei.clusts) <- colnames(bin.Adj)

### Visualizing the clusters
SC@kmeans$kpart<-lei.clusts
Factor<-factor(lei.clusts)
K<- length(levels(Factor))
plottSNE(SC)

# Silhouette plot
withr::with_options(repr.plot.width=12, repr.plot.height=25)
plotSilhouette(SC,K=K)

 # Jaccard Similarity
withr::with_options(repr.plot.width=10, repr.plot.height=12)
Jaccard(SC, K=K, plot = TRUE)   

lei.clusts <- leiden(bin.Adj, resolution_parameter = 1)   # Leiden algorithm
names(lei.clusts) <- colnames(bin.Adj)

### Visualizing the clusters
SC@kmeans$kpart<-lei.clusts
Factor<-factor(lei.clusts)
K<- length(levels(Factor))

withr::with_options(repr.plot.width=6, repr.plot.height=6)
plottSNE(SC)

# Silhouette plot
withr::with_options(repr.plot.width=12, repr.plot.height=25)
plotSilhouette(SC,K=K)

 # Jaccard Similarity
withr::with_options(repr.plot.width=10, repr.plot.height=12)
Jaccard(SC, K=K, plot = TRUE)