if(!file.exists("data/rawcounts_symbols.rda")){
  ### Load counts (from CellRanger) ----
  rawcounts_be2c<-read.csv(gzfile("data/BE2C.csv.gz"),as.is=TRUE,row.names=1)
  ncol(rawcounts_be2c) # 962 BE2C cells
  rawcounts_kelly<-read.csv(gzfile("data/Kelly.csv.gz"),as.is=TRUE,row.names=1)
  ncol(rawcounts_kelly) # 1105 Kelly cells
  # R loads "-" as ".", we must fix this
  colnames(rawcounts_be2c)<-gsub("\\.","-",colnames(rawcounts_be2c))
  colnames(rawcounts_be2c)<-paste0("be2c_",colnames(rawcounts_be2c))
  colnames(rawcounts_kelly)<-gsub("\\.","-",colnames(rawcounts_kelly))
  colnames(rawcounts_kelly)<-paste0("kelly_",colnames(rawcounts_kelly))
  # Merge the two datasets in a single raw counts matrix
  # rows are genes, columns are single cells
  rawcounts<-cbind(rawcounts_be2c,rawcounts_kelly)
  rawcounts<-as.matrix(rawcounts)
  dim(rawcounts) # 32738 ENSEMBL genes, 2067 samples
  #save(rawcounts,file="data/rawcounts_ensg.rda")
  
  ## Convert to gene symbols
  # Ensembl ids mapping to the same gene will see their raw counts summed up
  ensgmat<-rawcounts
  tmp<-ens2eg(rownames(ensgmat))
  convlist<-eg2sym(tmp)
  names(convlist)<-names(tmp)
  rawcounts<-squish(ensgmat,convlist=convlist,method="sum",verbose=TRUE)
  dim(rawcounts) # 22135 genes, 2067 samples
  save(rawcounts,file="data/rawcounts_symbols.rda")
  
  ## Save as CSV
  ksym<-rawcounts[,grep("kelly",colnames(rawcounts))]
  bsym<-rawcounts[,grep("be2c",colnames(rawcounts))]
  write.csv(bsym,file="results/Supp_File_S1_BE2C_rawcounts.csv")
  write.csv(ksym,file="results/Supp_File_S2_Kelly_rawcounts.csv")
} else {load("data/rawcounts_symbols.rda")}

## Normalization
# Extract gene lengths (more precisely, transcript lengths)
fname<-"data/genelengths.rda"
if(!file.exists(fname)){
  library(GenomicFeatures)
  supportedUCSCtables(genome="hg38", url="http://genome.ucsc.edu/cgi-bin/")
  hg<-makeTxDbFromUCSC(genome="hg38",tablename="refGene")
  exonic<-exonsBy(hg,by="gene")
  redexonic<-reduce(exonic)
  genelengths<-sum(width(redexonic))
  names(genelengths)<-eg2sym(names(genelengths))
  genelengths<-genelengths[!duplicated(names(genelengths))]
  genelengths<-genelengths[genelengths>0]
  genelengths<-genelengths[!is.na(genelengths)]
  save(genelengths,file=fname)
}else{load(fname)}

# Function to calculate FPKM (https://www.rna-seqblog.com/rpkm-fpkm-and-tpm-clearly-explained/)
fpkm<-function(counts,genelengths){
  common<-intersect(rownames(counts),names(genelengths))
  counts<-counts[common,]
  lengths<-genelengths[common]
  fpms<-apply(counts,2,function(x){1E6*x/sum(x)})
  fpkms<-fpms
  for(i in 1:nrow(fpms)){
    fpkms[i,]<-1E3*fpms[i,]/lengths[i]
  }
  return(fpkms)
}
fname<-"data/fpkms.rda"
if(!file.exists(fname)){
  fpkms<-fpkm(rawcounts,genelengths)
  save(fpkms,file=fname)
}
# Function to calculate TPM (https://www.rna-seqblog.com/rpkm-fpkm-and-tpm-clearly-explained/)
tpm<-function(counts,genelengths){
  common<-intersect(rownames(counts),names(genelengths))
  counts<-counts[common,]
  lengths<-genelengths[common]
  intermediate<-counts
  for(i in 1:nrow(counts)){
    intermediate[i,]<-1E3*counts[i,]/lengths[i]
  }
  tpms<-apply(intermediate,2,function(x){1E6*x/sum(x)})
  return(tpms)
}
fname<-"data/tpms.rda"
if(!file.exists(fname)){
  tpms<-tpm(rawcounts,genelengths)
  save(tpms,file=fname)
}

# Variance-Stabilizing Transformation (expmat)
fname<-"data/tpms.rda"
if(!file.exists(fname)){
  vstmat<-vst(rawcounts)
  save(vstmat,file=fname)
}

## Log-Normalize with Seurat
# By default, Seurat employs a global-scaling normalization method LogNormalize that
# normalizes the gene expression measurements for each cell by the total expression,
# multiplies this by a scale factor (10,000 by default), and log-transforms the result:
fname<-"data/expmat.rda"
if(!file.exists(fname)){
  seuset<-CreateSeuratObject(counts=rawcounts,project="nblcells",min.cells=3,min.features=1000)
  seuset # 15782 genes, 2067 samples
  seuset<-NormalizeData(seuset,normalization.method="LogNormalize",scale.factor=10000)
  save(seuset,file="data/seuset.rda")
  
  ## Mean variability plot showing most expressed and variable genes
  expmat<-as.matrix(seuset[["RNA"]]@data)
  dim(expmat) # 15782 genes, 2067 samples
  save(expmat,file=fname)
}

### Chromosome bands ----
fname<-"data/mlist.rda"
if(!file.exists(fname)){
  mdf<-msigdbr(species="Homo sapiens") # Retrieve all human gene sets
  mlist<-mdf %>% split(x=.$gene_symbol,f=.$gs_name)
  save(mlist,file=fname)
}else{load(fname)}
chrom_bands <- mlist[grep("chr",names(mlist))] 
chrs<-c("chr1","chr2","chr3","chr4","chr5","chr6","chr7","chr8","chr9","chr10","chr11","chr12","chr13",
        "chr14","chr15","chr16","chr17","chr18","chr19","chr20","chr21","chr22","chrX","chrY")

# Find gene
gene<-"MYCN"
for(i in 1:length(chrom_bands)){
  band<-names(chrom_bands)[i]
  genes_here<-chrom_bands[[band]]
  if(gene%in%genes_here){
    message(gene," is in ",band)
  }
}
# MYCN is in chr2p24


# Kelly
cols<-c()
pickcols<-c("royalblue4","skyblue")
means<-c()
coords<-c()
for(i in 1:length(chrs)){
  chr<-chrs[i]
  herecol<-pickcols[(i%%2)+1]
  sub<-chrom_bands[grep(paste0(chr,"(p|q)"),names(chrom_bands))]
  herecols<-rep(herecol,length(sub))
  cols<-c(cols,herecols)
  coords<-c(coords,rep(chr,length(sub)))
  for(band in names(sub)){
    genes_here<-sub[[band]]
    genes_here<-intersect(rownames(kmat),genes_here)
    allexp<-kmat[genes_here,]
    mean<-mean(allexp)
    means<-c(means,mean)
    names(means)[length(means)]<-band
  }
}

png("plots/003_chrom_kelly.png",w=11000,h=1000,res=300)
par(las=2)
bp<-barplot(means,col=cols,ylab="Mean expression in band (TPM)",xaxt="n",main="Kelly gene expression by chromosome band",ylim=c(0,650))
for(i in 1:length(chrs)){
  chr<-chrs[i]
  xwhere<-mean(bp[which(coords==chr)])
  text(xwhere,600,labels=chr,font=2)
}
axis(1,at=bp,labels=names(means),cex.axis=0.7)
dev.off()

# BE2C
cols<-c()
pickcols<-c("red3","salmon")
means<-c()
coords<-c()
for(i in 1:length(chrs)){
  chr<-chrs[i]
  herecol<-pickcols[(i%%2)+1]
  sub<-chrom_bands[grep(paste0(chr,"(p|q)"),names(chrom_bands))]
  herecols<-rep(herecol,length(sub))
  cols<-c(cols,herecols)
  coords<-c(coords,rep(chr,length(sub)))
  for(band in names(sub)){
    genes_here<-sub[[band]]
    genes_here<-intersect(rownames(bmat),genes_here)
    allexp<-bmat[genes_here,]
    mean<-mean(allexp)
    means<-c(means,mean)
    names(means)[length(means)]<-band
  }
}

png("plots/003_chrom_be2c.png",w=11000,h=1000,res=300)
par(las=2)
bp<-barplot(means,col=cols,ylab="Mean expression in band (TPM)",xaxt="n",main="BE2C gene expression by chromosome band",ylim=c(0,650))
for(i in 1:length(chrs)){
  chr<-chrs[i]
  xwhere<-mean(bp[which(coords==chr)])
  text(xwhere,600,labels=chr,font=2)
}
axis(1,at=bp,labels=names(means),cex.axis=0.7)
dev.off()

fname<-"data/harenza/rawcounts_symbols.rda"
if(!file.exists(fname)){
  ### Load counts (from CellRanger) ----
  rawcounts<-read.delim("data/harenza/harenza.counts.txt.gz",as.is=TRUE,skip=1,row.names=1)
  rawcounts<-as.matrix(rawcounts[,6:ncol(rawcounts)])
  colnames(rawcounts)<-gsub("\\.sorted\\.bam","",colnames(rawcounts))
  colnames(rawcounts)<-gsub("harenza_","",colnames(rawcounts))
  colnames(rawcounts)<-gsub("_.+","",colnames(rawcounts))
  colnames(rawcounts)<-gsub("\\.","-",colnames(rawcounts))
  save(rawcounts,file="data/harenza/rawcounts_ensg.rda")
  dim(rawcounts) # 58721 genes, 40 samples
  ## Convert to gene symbols
  ensgmat<-rawcounts
  rownames(ensgmat)<-gsub("\\..+","",rownames(ensgmat))
  tmp<-ens2eg(rownames(ensgmat))
  convlist<-eg2sym(tmp)
  names(convlist)<-names(tmp)
  rawcounts<-squish(ensgmat,convlist=convlist,method="sum",verbose=TRUE)
  dim(rawcounts) # 26131 genes, 40 samples
  save(rawcounts,file=fname)
} else {load(fname)}


### Normalization ----
# Extract gene lengths (more precisely, transcript lengths)
fname<-"data/genelengths.rda"
if(!file.exists(fname)){
  library(GenomicFeatures)
  supportedUCSCtables(genome="hg38", url="http://genome.ucsc.edu/cgi-bin/")
  hg<-makeTxDbFromUCSC(genome="hg38",tablename="refGene")
  exonic<-exonsBy(hg,by="gene")
  redexonic<-reduce(exonic)
  genelengths<-sum(width(redexonic))
  names(genelengths)<-eg2sym(names(genelengths))
  genelengths<-genelengths[!duplicated(names(genelengths))]
  genelengths<-genelengths[genelengths>0]
  genelengths<-genelengths[!is.na(genelengths)]
  save(genelengths,file=fname)
}else{load(fname)}



# Function to calculate FPKM (https://www.rna-seqblog.com/rpkm-fpkm-and-tpm-clearly-explained/)
fpkm<-function(counts,genelengths){
  common<-intersect(rownames(counts),names(genelengths))
  counts<-counts[common,]
  lengths<-genelengths[common]
  fpms<-apply(counts,2,function(x){1E6*x/sum(x)})
  fpkms<-fpms
  for(i in 1:nrow(fpms)){
    fpkms[i,]<-1E3*fpms[i,]/lengths[i]
  }
  return(fpkms)
}
fname<-"data/harenza/fpkms.rda"
if(!file.exists(fname)){
  fpkms<-fpkm(rawcounts,genelengths)
  save(fpkms,file=fname)
}
# Function to calculate TPM (https://www.rna-seqblog.com/rpkm-fpkm-and-tpm-clearly-explained/)
tpm<-function(counts,genelengths){
  common<-intersect(rownames(counts),names(genelengths))
  counts<-counts[common,]
  lengths<-genelengths[common]
  intermediate<-counts
  for(i in 1:nrow(counts)){
    intermediate[i,]<-1E3*counts[i,]/lengths[i]
  }
  tpms<-apply(intermediate,2,function(x){1E6*x/sum(x)})
  return(tpms)
}
fname<-"data/harenza/tpms.rda"
if(!file.exists(fname)){
  tpms<-tpm(rawcounts,genelengths)
  save(tpms,file=fname)
}

# Variance-Stabilizing Transformation (expmat)
fname<-"data/harenza/vstmat.rda"
if(!file.exists(fname)){
  vstmat<-vst(rawcounts)
  save(vstmat,file=fname)
}

### Load data ----
# Harenza
load("data/harenza/tpms.rda")
harenza<-tpms
# Our own
load("data/tpms.rda")
data<-tpms
bmat<-data[,grep("be2c",colnames(data))]
kmat<-data[,grep("kelly",colnames(data))]
b<-apply(bmat,1,sum)
k<-apply(kmat,1,sum)
# Correct names
colnames(harenza)[colnames(harenza)=="SK-N-BE-2--C"]<-"BE2C"
colnames(harenza)[colnames(harenza)=="SK-N-BE-2-"]<-"BE2"





### Correlation matrices ----
output<-matrix(NA,nrow=ncol(harenza),ncol=2)
colnames(output)<-c("SCC","SCC p-value")
rownames(output)<-colnames(harenza)

# Kelly
grpm<-k
for (i in 1:ncol(harenza)){
    cellname<-colnames(harenza)[i]
    message("Doing ",cellname)
    hrpm<-harenza[,cellname]
    common<-intersect(names(grpm),names(hrpm))
    grpm<-grpm[common]
    hrpm<-hrpm[common]
    x<-grpm
    y<-hrpm
    
    ii<-1
    for(cortype in c("s")){
        cortest<-cor.test(x,y,method=cortype)
        coeff<-signif(cortest$estimate,4)
        p<-signif(cortest$p.value,3)
        output[i,ii]<-coeff
        ii<-ii+1
        output[i,ii]<-p
        ii<-ii+1
    }
}
output<-output[order(-output[,1]),]

pander(output,style="rmarkdown") # Correlation between Kelly single cell dataset and other datasets

#png("plots/005_cortests_kelly_vs_harenza.png",w=2000,h=5000,res=350)
#grid.newpage()
#grid.table(output,theme=ttheme_default(base_colour="navy"))
#dev.off()
write.xlsx2(output,file="results/Supp_Table_S2_Kelly_vs_Bulk.xlsx")

# BE2C
output<-matrix(NA,nrow=ncol(harenza),ncol=2)
colnames(output)<-c("SCC","SCC p-value")
rownames(output)<-colnames(harenza)
grpm<-b
for (i in 1:ncol(harenza)){
    cellname<-colnames(harenza)[i]
    message("Doing ",cellname)
    hrpm<-harenza[,cellname]
    common<-intersect(names(grpm),names(hrpm))
    grpm<-grpm[common]
    hrpm<-hrpm[common]
    x<-grpm
    y<-hrpm
    
    ii<-1
    for(cortype in c("s")){
        cortest<-cor.test(x,y,method=cortype)
        coeff<-signif(cortest$estimate,4)
        p<-signif(cortest$p.value,3)
        output[i,ii]<-coeff
        ii<-ii+1
        output[i,ii]<-p
        ii<-ii+1
    }
}
output<-output[order(-output[,1]),]

pander(output,style="rmarkdown") # Correlation between BE2C single cell dataset and other datasets

#png("plots/005_cortests_be2c_vs_harenza.png",w=2000,h=5000,res=350)
#grid.table(output,theme=ttheme_default(base_colour="red3"))
#dev.off()
write.xlsx2(output,file="results/Supp_Table_S1_BE2C_vs_Bulk.xlsx")

### Scatter plots bulk vs sc ----
pseudo<-0.0001
toshow<-c("MYCN","GAPDH","GUSB","ACTB","B2M")
# Kelly
x<-log10(k+pseudo)
y<-log10(harenza[,"KELLY"]+pseudo)
#png("plots/005_scatter_kelly.png",w=4000,h=3000,res=600)
scatter(x,y,xlab="Single Cell RNA-Seq (Sum of TPMs)",ylab="Bulk RNA-Seq (TPM)",main="Kelly cells",col=col_kelly,method="spearman")
textplot3(x[toshow],y[toshow],words=toshow,font=2)

#dev.off()

# BE2C
x<-log10(b+pseudo)
y<-log10(harenza[,"BE2C"]+pseudo)

#png("plots/005_scatter_be2c.png",w=4000,h=3000,res=600)
scatter(x,y,xlab="Single Cell RNA-Seq (Sum of TPMs)",ylab="Bulk RNA-Seq (TPM)",main="BE2C cells",col=col_be2c,method="spearman")
textplot3(x[toshow],y[toshow],words=toshow,font=2)

#dev.off()

### Rtsne ----
# Load cell line annotation
cl<-read.csv("data/NBLcellLines.csv",header=TRUE)
cl<-setNames(cl[,4],cl[,1])

# Everything together
lh<-log10(harenza+pseudo)
lk<-log10(k+pseudo)
lb<-log10(b+pseudo)
common<-intersect(rownames(lh),intersect(names(lk),names(lb)))
tsnemat<-cbind(lh[common,],lk[common],lb[common])
colnames(tsnemat)[(ncol(tsnemat)-1):ncol(tsnemat)]<-c("scKELLY","scBE2C")
# Prepare the matrix
topvars<-names(sort(apply(tsnemat,1,var),decreasing=TRUE))[1:5000]
tsnemat<-tsnemat[topvars,]
# Seed for TSNE and calculate TSNE
set.seed(4)
ttt<-Rtsne(t(tsnemat),perplexity=10)


# Start preparing the plot
x<-setNames(ttt$Y[,1],colnames(tsnemat))
y<-setNames(ttt$Y[,2],colnames(tsnemat))
# Shapes
shapes<-rep(15,length(x))
names(shapes)<-names(x)
shapes[names(cl[cl=="notAmplified"])]<-16
#
#png("plots/005_tsne_with_bulk.png",w=3000,h=3000,res=500)
plot(x,y,pch=shapes,xlab="TSNE1",ylab="TSNE2",main="TSNE Representation of NBL Cell Lines",
     xlim=1.1*c(min(x),max(x)),ylim=1.1*c(min(y),max(y)),
     col=c(rep("#00000099",length(x)-2),col_kelly,col_be2c)
)
set.seed(4)
textplot3(x,y,words=names(x),cex=0.8,
          font=c(rep(1,length(x)-2),2,2),
          col=c(rep("black",length(x)-2),col_kelly,col_be2c),padding="  ",pos=3,offset=0.1
)
legend("bottomleft",pch=c(15,16),c("Amplified","Not Amplified"),title="MYCN status")

#dev.off()

# Again without BE2C
x<-setNames(ttt$Y[,1],colnames(tsnemat))
y<-setNames(ttt$Y[,2],colnames(tsnemat))
x<-x[1:(length(x)-1)]
y<-y[1:(length(y)-1)]
# Shapes
shapes<-rep(15,length(x))
names(shapes)<-names(x)
shapes[names(cl[cl=="notAmplified"])]<-16
#
#png("plots/005_tsne_with_bulk_Kelly.png",w=6000,h=3000,res=600)
plot(x,y,pch=shapes,xlab="TSNE1",ylab="TSNE2",main="Clustering of NBL Cell Lines",
     xlim=1.1*c(min(x),max(x)),ylim=1.1*c(min(y),max(y)),
     col=c(rep("#00000099",length(x)-1),col_kelly)
)
set.seed(4)
textplot3(x,y,words=names(x),cex=0.8,
          font=c(rep(1,length(x)-1),2),
          col=c(rep("black",length(x)-1),col_kelly),padding="  ",pos=3,offset=0.1
)
legend("bottomleft",pch=c(15,16),c("Amplified","Not Amplified"),title="MYCN status")

#dev.off()

### Effects of cell cycle and read numbers on clustering ----
# Cell Cycle Markers, from Tirosh et al, 2015
ccgenes<-readLines("data/regev_lab_cell_cycle_genes.txt")
ccgenes<-eg2sym(sym2eg(ccgenes))
s.genes <- ccgenes[1:43]
g2m.genes <- ccgenes[44:97]
s.genes<-intersect(s.genes,rownames(seuset))
g2m.genes<-intersect(g2m.genes,rownames(seuset))
# Apply scoring
seuset<-CellCycleScoring(seuset,s.features=s.genes,g2m.features=g2m.genes,set.ident=TRUE)

# Rtsne, top var genes
allvars<-apply(expmat,1,var)
topvars<-names(sort(allvars,dec=TRUE))[1:1000]
topvarmat<-expmat[topvars,]
fname<-"data/rtsne.rda"
if(!file.exists(fname)){
    set.seed(1)
    rtsne<-Rtsne(t(topvarmat))
    save(rtsne,file=fname)
}else{load(fname)}

# Coloring cell lines
x<-setNames(rtsne$Y[,1],colnames(topvarmat))
y<-setNames(rtsne$Y[,2],colnames(topvarmat))
mycols<-rep("black",ncol(topvarmat))
mycols[grep("be2c",colnames(topvarmat))]<-col_be2c
mycols[grep("kelly",colnames(topvarmat))]<-col_kelly
#png("plots/006b_tsne_seurat_celllines.png",w=3000,h=3000,res=600)
plot(x,y,pch=20,xlab="TSNE1",ylab="TSNE2",main="TSNE clustering of KELLY and BE2C",col=mycols,xlim=c(min(x),max(x)*1.5))
mtext(paste0(length(x)," cells"),cex=0.8,font=2)
grid()
legend("bottomright",pch=19,col=c(col_be2c,col_kelly),legend=c("BE2C","Kelly"))

#dev.off()

# Color cell cycle phases
phases<-setNames(as.character(seuset@meta.data$Phase),rownames(seuset@meta.data))
mycols<-phases
mycols[phases=="G1"]<-"salmon"
mycols[phases=="S"]<-"cornflowerblue"
mycols[phases=="G2M"]<-"seagreen"
mycols<-mycols[colnames(topvarmat)]
#png("plots/006b_tsne_seurat_celllcycle.png",w=3000,h=3000,res=600)
plot(x,y,pch=20,xlab="TSNE1",ylab="TSNE2",main="TSNE clustering of KELLY and BE2C",col=mycols,xlim=c(min(x),max(x)*1.5))
mtext(paste0(length(x)," cells"),cex=0.8,font=2)
grid()
legend("bottomright",pch=19,col=c("salmon","cornflowerblue","seagreen"),legend=c("G1","S","G2M"))

#dev.off()

# Barplot cell cycle distribution
kphases<-phases[grep("kelly",names(phases))]
bphases<-phases[grep("be2c",names(phases))]
ktab<-table(kphases)[c("G1","S","G2M")]
btab<-table(bphases)[c("G1","S","G2M")]

#png("plots/006b_phases_barplots.png",w=3000,h=3000,res=500)
par(mfrow=c(1,2))
max<-max(c(ktab,btab))*1.2
bp<-barplot(ktab,col=c("salmon","cornflowerblue","seagreen"),main="Kelly",ylim=c(0,max),ylab="nr. cells")
mtext(paste0(length(kphases)," cells"),cex=0.8)
perc<-round(100*ktab/length(kphases),2)
text(bp,ktab,labels=paste0(perc,"%"),pos=3,font=3)
bp<-barplot(btab,col=c("salmon","cornflowerblue","seagreen"),main="BE2C",ylim=c(0,max))
mtext(paste0(length(bphases)," cells"),cex=0.8)
perc<-round(100*btab/length(bphases),2)
text(bp,btab,labels=paste0(perc,"%"),pos=3,font=3)

#dev.off()
par(mfrow=c(1,1))

# Color nr. reads coloring
nreads<-apply(rawcounts,2,sum)
colfunc <- colorRampPalette(c("cornflowerblue","red3","orange"))
mycols<-colfunc(100)[as.numeric(cut(nreads,breaks=100))]
#png("plots/006b_tsne_seurat_nreads.png",w=3000,h=3000,res=600)
layout(matrix(1:2,ncol=2), width = c(3,1),height = c(1,1))
par(mar=c(5.1,4.1,4.1,0.1))
plot(x,y,pch=20,xlab="TSNE1",ylab="TSNE2",main="TSNE clustering",col=mycols,xlim=c(min(x),max(x)*1.5))
mtext(paste0(length(x)," cells"),cex=0.8,font=2)
grid()
par(mar=c(5.1,0.1,4.1,0.1))
plot(c(0,2),c(0,1),type='n',axes=F,xlab='',ylab='',main='Nr. Reads (k)')
legend_image<-as.raster(rev(matrix(colfunc(100), ncol=1)))
rasterImage(legend_image,0,0,1,1)
text(x=1.5,y=seq(0,1,l=5),labels=round(quantile(nreads/1E3)))

#dev.off()

# Read counts distributions
#png("plots/006b_nreads_lineplots.png",w=3000,h=3000,res=600)
knreads<-nreads[grep("kelly",names(nreads))]
bnreads<-nreads[grep("be2c",names(nreads))]
plot(density(knreads),col=col_kelly,lwd=3,ylim=c(0,4e-5),main="Nr. reads/cell distribution")
lines(density(bnreads),col=col_be2c,lwd=3)
legend("topright",legend=c(
    paste0("Kelly (mean=",round(mean(knreads),2),")"),
    paste0("BE2C (mean=",round(mean(bnreads),2),")")
),lwd=3,col=c(col_kelly,col_be2c))

#dev.off()

### Regress out cell cycle and nUMI ----
fname<-"data/seuset_regressed.rda"
if(!file.exists(fname)){
    seuset<-ScaleData(object=seuset,vars.to.regress=c("nUMI","S.Score","G2M.Score"))
    save(seuset,file=fname)
}else{load(fname)}
regmat<-as.matrix(seuset[["RNA"]]@scale.data)
dim(regmat) # 15782 genes 2067 cells

save(regmat,file="data/regmat.rda")
topvarmat<-regmat

# Rtsne, top var genes
allvars<-apply(topvarmat,1,var)
topvars<-names(sort(allvars,dec=TRUE))[1:1000]
topvarmat<-topvarmat[topvars,]
set.seed(1)
rtsne<-Rtsne(t(topvarmat))
save(rtsne,file=fname)

# Coloring cell lines
x<-setNames(rtsne$Y[,1],colnames(topvarmat))
y<-setNames(rtsne$Y[,2],colnames(topvarmat))
mycols<-rep("black",ncol(topvarmat))
mycols[grep("be2c",colnames(topvarmat))]<-col_be2c
mycols[grep("kelly",colnames(topvarmat))]<-col_kelly
#png("plots/006c_tsne_seurat_celllines_postregr.png",w=3000,h=3000,res=600)
plot(x,y,pch=20,xlab="TSNE1",ylab="TSNE2",main="Post-regression TSNE",col=mycols,xlim=c(min(x),max(x)*1.5))
mtext(paste0(length(x)," cells"),cex=0.8,font=2)
grid()
legend("bottomright",pch=19,col=c(col_be2c,col_kelly),legend=c("BE2C","Kelly"))

#dev.off()

# Color cell cycle phases
phases<-setNames(as.character(seuset@meta.data$Phase),rownames(seuset@meta.data))
mycols<-phases
mycols[phases=="G1"]<-"salmon"
mycols[phases=="S"]<-"cornflowerblue"
mycols[phases=="G2M"]<-"seagreen"
mycols<-mycols[colnames(topvarmat)]
#png("plots/006c_tsne_seurat_celllcycle_postregr.png",w=3000,h=3000,res=600)
plot(x,y,pch=20,xlab="TSNE1",ylab="TSNE2",main="Post-regression TSNE",col=mycols,xlim=c(min(x),max(x)*1.5))
mtext(paste0(length(x)," cells"),cex=0.8,font=2)
grid()
legend("bottomright",pch=19,col=c("salmon","cornflowerblue","seagreen"),legend=c("G1","S","G2M"))

#dev.off()

# Color nr. reads coloring
nreads<-apply(rawcounts,2,sum)
colfunc <- colorRampPalette(c("cornflowerblue","red3","orange"))
mycols<-colfunc(100)[as.numeric(cut(nreads,breaks=100))]
#png("plots/006c_tsne_seurat_nreads_postregr.png",w=3000,h=3000,res=600)
layout(matrix(1:2,ncol=2), width = c(2,1),height = c(1,1))
plot(x,y,pch=20,xlab="TSNE1",ylab="TSNE2",main="Post-regression TSNE",col=mycols,xlim=c(min(x),max(x)*1.5))
mtext(paste0(length(x)," cells"),cex=0.8,font=2)
grid()
plot(c(0,2),c(0,1),type='n',axes=F,xlab='',ylab='',main='Nr. Reads (k)')
legend_image<-as.raster(rev(matrix(colfunc(100), ncol=1)))
rasterImage(legend_image,0,0,1,1)
text(x=1.5,y=seq(0,1,l=5),labels=round(quantile(nreads/1E3)))

#dev.off()

### Bulk signature (from Harenza) ----
# We must compare log2FC, since the bulk RNA-Seq has only a single sample per condition
load("data/harenza/vstmat.rda")
hsig<-log2(vstmat[,"SK-N-BE-2--C"]/vstmat[,"KELLY"])

load("data/expmat.rda")
min(expmat[expmat!=0]) # 0.05

expmat<-expmat+0.01
scb<-apply(expmat[,grep("be2c",colnames(expmat))],1,mean)
sck<-apply(expmat[,grep("kelly",colnames(expmat))],1,mean)
scsig<-log2(scb/sck)

#png("plots/007_sc_vs_bulk.png",w=3000,h=3000,res=600)
scatter(hsig,scsig,xlab="log2FC, Bulk RNA-Seq",ylab="log2FC, Single Cell RNA-Seq",main="BE2C vs. Kelly Differential Expression")

#dev.off()

### Volcano plot ----
load("results/007_res_be2c_vs_kelly.rda")
x<-setNames(res$log2fc,rownames(res))
y<-setNames(-log10(res$fdr),rownames(res))


#png("plots/007_volcanoes.png",w=6000,h=3000,res=600)
par(mfrow=c(3,2))
pthrs<-c(2,5,10,20,50,100)
toshow<-c("RPS25","RPL27","MYCN")
for(pthr in pthrs){
    plot(x,y,pch=20,xlab="log2 Fold Change",ylab="-log10(FDR)",col="#00000011",main="BE2C vs. Kelly")
    up<-which(x>0&y>pthr)
    points(x[up],y[up],col="#FF000011",pch=20)
    dn<-which(x<0&y>pthr)
    points(x[dn],y[dn],col="#0000FF11",pch=20)
    mtext(paste0("Significant at FDR=10^-",pthr,": ",length(up),"(up), ",length(dn),"(down)"),cex=0.6)
    if(pthr==2){
        text(x[toshow],y[toshow],labels=toshow,font=2)
    }
}

#dev.off()

fname<-"data/mlist.rda"
if(!file.exists(fname)){
  mdf<-msigdbr(species="Homo sapiens") # Retrieve all human gene sets
  mlist<-mdf %>% split(x=.$gene_symbol,f=.$gs_name)
  save(mlist,file=fname)
}else{load(fname)}


### Get signature BE2C vs. Kelly ----
load("results/007_res_be2c_vs_kelly.rda")
signature<-setNames(-log10(res$p)*sign(res$log2fc),rownames(res))

### Check contrast orientation ----
signature[1:10]

load("data/tpms.rda")
mean(tpms["RPS25",grep("be2c",colnames(tpms))])

mean(tpms["RPS25",grep("kelly",colnames(tpms))])

mean(tpms["RPL27",grep("be2c",colnames(tpms))])

mean(tpms["RPL27",grep("kelly",colnames(tpms))])

# Positive is, as planned, higher in BE2C


### Run GSEA (fgsea function) ----
fname<-"results/008_gsea.rda"
if(!file.exists(fname)){
    library(fgsea)
    gsea<-fgsea(pathways=mlist,stats=signature,nperm=1E6,minSize=4,maxSize=Inf,nproc=7)
    gsea<-gsea[order(gsea$pval),]
    gsea<-as.data.frame(gsea)
    save(gsea,file=fname)
    # Save as supp table
    write.csv(gsea,file="results/008_gsea_BE2C_vs_Kelly.csv")
}else{load(fname)}

### Run GSEA (corto function) ----
fname<-"results/008_gsea_corto.rda"
if(!file.exists(fname)){
    gsea<-matrix(nrow=length(mlist),ncol=3)
    rownames(gsea)<-names(mlist)
    colnames(gsea)<-c("ES","NES","p")
    set.seed(1)
    pb<-txtProgressBar(0,length(mlist),style=3)
    for(i in 1:length(mlist)){
        pname<-names(mlist)[i]
        pathway<-mlist[[pname]]
        if(length(intersect(names(signature),pathway))>1){
            obj<-gsea(signature,pathway,method="pareto")
            gsea[pname,]<-c(obj$es,obj$nes,obj$p)
        }else{
            gsea[pname,]<-c(0,0,1)
        }
        setTxtProgressBar(pb,i)
    }
    fdr<-p.adjust(gsea[,"p"],method="BH")
    gsea<-cbind(gsea,fdr)
    colnames(gsea)[ncol(gsea)]<-"fdr"
    save(gsea,file=fname)
    # Save as supp table
    write.csv(gsea,file="results/008_gsea_BE2C_vs_Kelly_corto.csv")
}else{load(fname)}

# All pathways
gsea<-as.data.frame(gsea)
gsea<-gsea[order(-abs(gsea$NES)),]


### Table of top pathways ----
top<-gsea[gsea$NES<0,][1:15,]
top<-rbind(top,gsea[gsea$NES>0,][15:1,])
top<-top[order(top$NES),]
toplot<-setNames(top$NES,rownames(top))
# Format
#names(toplot)<-gsub("GO_","",names(toplot))
names(toplot)<-gsub("_"," ",names(toplot))
names(toplot)<-str_to_title(names(toplot))
#png("plots/008_gsea_be2c_vs_kelly.png",w=6000,h=3000,res=500)
par(mar=c(4,1,3,1))
bp<-barplot(toplot,horiz=TRUE,xlab="Normalized Enrichment Score",
            xlim=1.3*c(-max(abs(toplot)),max(abs(toplot))),
            main="BE2C vs. Kelly, top Pathways",
            col=rep(c("cornflowerblue","salmon"),each=15),
            yaxt="n",cex.main=2
)
text(0,bp[1:15,1],names(toplot)[1:15],pos=4)
text(0,bp[16:30,1],names(toplot)[16:30],pos=2)
abline(v=c(-p2z(0.05),p2z(0.05)),lty=2)

#dev.off()            

### Check for Neuroblastoma ----
gsea[grep("NEUROBLASTOMA|NBL",rownames(gsea)),]

### Plot top GSEAs ----
pathways<-c("REACTOME_METABOLISM_OF_RNA","CHICAS_RB1_TARGETS_CONFLUENT","GO_RIBOSOME_BIOGENESIS","LASTOWSKA_NEUROBLASTOMA_COPY_NUMBER_UP","LASTOWSKA_NEUROBLASTOMA_COPY_NUMBER_DN")
for(pathway in pathways){
    set<-mlist[[pathway]]
    obj<-gsea(signature,set,method="pareto")
    #png(paste0("plots/008_gsea_",pathway,".png"),w=3000,h=3000,res=600)
    plot_gsea(obj,ext_nes=gsea[rownames(gsea)==pathway,"NES"],title=pathway,colBarcode = "#00000033")
    #dev.off()
}

## Double gsea (TEST)
path1<-"LASTOWSKA_NEUROBLASTOMA_COPY_NUMBER_UP"
path2<-"LASTOWSKA_NEUROBLASTOMA_COPY_NUMBER_DN"
set1<-mlist[[path1]]
set2<-mlist[[path2]]
obj<-gsea2(signature,set1,set2,method="pareto")
plot_gsea2(obj)

load("data/seuset_regressed.rda")


### Calculate Rtsne ----
load("data/rtsne.rda")
load("data/regmat.rda")
x<-setNames(rtsne$Y[,1],colnames(regmat))
y<-setNames(rtsne$Y[,2],colnames(regmat))

### Overlay single genes over Rtsne ----
# Get TPM per cell
load("data/tpms.rda")
genes<-c("MYCN","GAPDH","ACTB","GUSB","B2M","ALK","LMO1")
for(gene in genes){
    vector<-rank(tpms[gene,])
    
    ## Color by TPM rank
    colfunc<-colorRampPalette(c("navy","grey","red3"))
    mycols<-colfunc(1000)[as.numeric(cut(vector,breaks=1000))]
    # Plot
    #png(paste0("plots/009_tsne_rank_",gene,".png"),w=4000,h=2000,res=450)
    layout(matrix(1:2,ncol=2), width = c(2,1),height = c(1,1))
    plot(x,y,pch=20,xlab="TSNE1",ylab="TSNE2",main=paste0(gene),col=mycols,xlim=c(min(x),max(x)*1.5))
    mtext(paste0(length(x)," cells"),cex=0.8,font=2)
    grid()
    plot(c(0,2),c(0,1),type='n',axes=F,xlab='',ylab='',main=paste0(" expression (rank)"))
    legend_image<-as.raster(rev(matrix(colfunc(1000), ncol=1)))
    rasterImage(legend_image,0,0,1,1)
    text(x=1.5,y=seq(0,1,l=5),labels=round(quantile(vector)))
    #dev.off()
}

for(gene in genes){
    vector<-rank(tpms[gene,])
    
    ## Color by TPM rank
    colfunc<-colorRampPalette(c("navy","grey","red3"))
    mycols<-colfunc(1000)[as.numeric(cut(vector,breaks=1000))]

    ## Color by TPM
    colfunc<-colorRampPalette(c("navy","navy","navy","grey","salmon","red","red3"))
    mycols<-colfunc(1000)[as.numeric(cut(vector,breaks=1000))]
    vector<-tpms[gene,]
    pseudo<-min(vector[vector!=0]) # Pseudocount to ignore dropout effects
    vector<-log10(vector+pseudo)
    
    # Plot
    #png(paste0("plots/009_tsne_log10_",gene,".png"),w=4500,h=2000,res=500)
    lmatrix<-t(c(1,2,2,2,3))
    layout(lmatrix)
    par(mar=c(5.1,4.1,4.1,2.1))
    hist(vector,main=paste0(gene,""),xlab="Log10 TPM")
    mycols<-colfunc(1000)[as.numeric(cut(vector,breaks=1000))]
    plot(x,y,pch=20,xlab="TSNE1",ylab="TSNE2",main=paste0(gene),col=mycols,xlim=c(min(x),max(x)*1.5))
    mtext(paste0(length(x)," cells"),cex=0.8,font=2)
    par(mar=c(5.1,1.1,4.1,2.1))
    plot(c(0,2),c(0,1),type='n',axes=F,xlab='',ylab='',main=paste0("Log10 TPM"))
    legend_image<-as.raster(rev(matrix(colfunc(1000), ncol=1)))
    rasterImage(legend_image,0,0,1,1)
    text(x=1.5,y=seq(0,1,l=5),labels=round(seq(min(vector),max(vector),length.out=5),2))
    #dev.off()
}

### Load the msigdb database 7.1.1 ----
fname<-"data/mlist.rda"
if(!file.exists(fname)){
  mdf<-msigdbr(species="Homo sapiens") # Retrieve all human gene sets
  mlist<-mdf %>% split(x=.$gene_symbol,f=.$gs_name)
  save(mlist,file=fname)
}else{load(fname)}


### Calculate single-cell gsea (better to use area) ----
fname<-"results/009_ssgsea_regexpmat.rda"
load("data/regmat.rda") # LogNormalized by Seurat + Regressed out cycle and nUMI
# regmat<-regmat+abs(min(regmat))
if(!file.exists(fname)){
    scalemat<-t(scale(t(regmat)))
    ssgsea<-area(signatures=scalemat,groups=mlist,minsize=5)
    save(ssgsea,file=fname)    
}else{load(fname)}

### Extract highest significance + highest variance pathways ----
dim(ssgsea) # 24472 pathways, 2067 cells

sumval<-apply(ssgsea,1,sum)
sumabs<-apply(ssgsea,1,function(x){sum(abs(x))})
var<-apply(ssgsea,1,var)

#png("plots/009b_ss_pathways.png",w=6000,h=2000,res=600)
par(mfrow=c(1,3))
plot(sumval,var,xlab="NES sum",ylab="NES variance")
plot(sumabs,var,xlab="NES sum of absolute",ylab="NES variance",main="Pathways in Single Cell Dataset")
toshow<-names(sort(var,dec=TRUE))[1:4]
text(sumabs[toshow],var[toshow],labels=toshow,pos=2,cex=0.4)
plot(sumabs,var/sumabs,xlab="NES sum of absolute",ylab="NES variance / sum of abs")
toshow<-names(sort(var/sumabs,dec=TRUE))[1:4]
text(sumabs[toshow],(var/sumabs)[toshow],labels=toshow,pos=2,cex=0.4)

#dev.off()

### Highest variable within Kelly and BE2C ----
ssk<-ssgsea[,grep("kelly",colnames(ssgsea))]
ssb<-ssgsea[,grep("be2c",colnames(ssgsea))]
# Kelly
sumval<-apply(ssk,1,sum)
sumabs<-apply(ssk,1,function(x){sum(abs(x))})
var<-apply(ssk,1,var)
#png("plots/009b_ss_pathways_kelly.png",w=6000,h=2000,res=600)
par(mfrow=c(1,3))
plot(sumval,var,xlab="NES sum",ylab="NES variance")
plot(sumabs,var,xlab="NES sum of absolute",ylab="NES variance",main="Pathways in Kelly Cells")
toshow<-names(sort(var,dec=TRUE))[1:6]
text(sumabs[toshow],var[toshow],labels=toshow,pos=2,cex=0.4)
plot(sumabs,var/sumabs,xlab="NES sum of absolute",ylab="NES variance / sum of abs")
toshow<-names(sort(var/sumabs,dec=TRUE))[1:4]
text(sumabs[toshow],(var/sumabs)[toshow],labels=toshow,pos=2,cex=0.4)

#dev.off()

# BE2C
sumval<-apply(ssb,1,sum)
sumabs<-apply(ssb,1,function(x){sum(abs(x))})
var<-apply(ssb,1,var)
#png("plots/009b_ss_pathways_be2c.png",w=6000,h=2000,res=600)
par(mfrow=c(1,3))
plot(sumval,var,xlab="NES sum",ylab="NES variance")
plot(sumabs,var,xlab="NES sum of absolute",ylab="NES variance",main="Pathways in BE2C Cells")
toshow<-names(sort(var,dec=TRUE))[1:6]
text(sumabs[toshow],var[toshow],labels=toshow,pos=2,cex=0.4)
plot(sumabs,var/sumabs,xlab="NES sum of absolute",ylab="NES variance / sum of abs")
toshow<-names(sort(var/sumabs,dec=TRUE))[c(1,3,4)]
text(sumabs[toshow],(var/sumabs)[toshow],labels=toshow,pos=4,cex=0.4)

#dev.off()

### Overlay selected pathways over TSNE ----
colfunc<-colorRampPalette(c("navy","navy","navy","grey","salmon","red","red3"))
mycols<-colfunc(1000)[as.numeric(cut(vector,breaks=1000))]
pathways<-c("KEGG_RIBOSOME","MORF_SOD1","HALLMARK_MYC_TARGETS_V1","GO_OXIDATIVE_PHOSPHORYLATION")
for(pathway in pathways){
    vector<-ssgsea[pathway,]
    # Plot
    #png(paste0("plots/009b_ss_pathway_",pathway,".png"),w=4000,h=2000,res=450)
    layout(matrix(1:2,ncol=2), width = c(2,1),height = c(1,1))
    mycols<-colfunc(1000)[as.numeric(cut(vector,breaks=1000))]
    plot(x,y,pch=20,xlab="TSNE1",ylab="TSNE2",main=paste0(pathway),col=mycols,xlim=c(min(x),max(x)*1.5))
    mtext(paste0(length(x)," cells"),cex=0.8,font=2)
    plot(c(0,2),c(0,1),type='n',axes=F,xlab='',ylab='',main=paste0("Normalized Enrichment Score"),cex.main=0.8)
    legend_image<-as.raster(rev(matrix(colfunc(1000), ncol=1)))
    rasterImage(legend_image,0,0,1,1)
    text(x=1.5,y=seq(0,1,l=5),labels=round(seq(min(vector),max(vector),length.out=5),2))
    #dev.off()
}

### Bonus: correlation MYCN vs. MYC pathway ----
path<-ssgsea["HALLMARK_MYC_TARGETS_V1",]
gene<-tpms["MYCN",]
cols<-rep("white",ncol(tpms))
cols[grep("kelly",colnames(tpms))]<-col_kelly
cols[grep("be2c",colnames(tpms))]<-col_be2c
#png("plots/009c_MYCN_vs_MYCpathway.png",w=3000,h=3000,res=600)
scatter(path,gene,col=cols,xlab="MSIGDB Hallmark MYC V1 Targets (NES)",ylab="MYCN Expression (TPM)",
        main="MYCN vs. MYC pathway"
)
legend("topright",pch=19,col=c(col_be2c,col_kelly),legend=c("BE2C","Kelly"))

#dev.off()