SCENIC/pySCENIC分析补充+小鼠数据示例2022-12-14

适用背景

之前写了两篇博客（四步完成单细胞数据调控网络流程分析-SCENIC/pySCENIC-2022-09-06和SCENIC/pySCENIC结果可视化 2022-11-08）介绍SCENIC/pySCENIC的使用，最近在使用的时候遇到一些问题，因此这篇文章作为补充，如果看不懂本篇可以查看前两篇博客。

补充内容主要有以下几点：

• 1、小鼠的数据库构建
• 2、从四步流程缩减到三步
• 3、环境构建遇到的一些errors
• 4、SCENIC版本的bugs

---

小鼠的数据库构建

之前的博客构建的是人的数据库，但博主最近分析需要用到小鼠的，因此需要构建一下新的数据库。正如之前博客写到的，其实构建这个数据库只需要替换3个文件，转录因子（TF）列表文件，feather文件已经tbl文件，建议使用最新的v10版本，当然也可以根据网站提示挑选适合自己数据的文件。我下载的文件如下：

• 小鼠TF文件：allTFs_mm.txt
  https://resources.aertslab.org/cistarget/tf_lists/
• feather文件：mm10_10kbp_up_10kbp_down_full_tx_v10_clust.genes_vs_motifs.rankings.feather
  https://resources.aertslab.org/cistarget/databases/mus_musculus/mm10/refseq_r80/mc_v10_clust/gene_based/
• tbl文件：motifs-v10nr_clust-nr.mgi-m0.001-o0.0.tbl.txt
  https://resources.aertslab.org/cistarget/motif2tf/motifs-v10nr_clust-nr.mgi-m0.001-o0.0.tbl

---

从四步流程缩减到三步

之前生成pyscenic需要的loom文件要先用R导出矩阵，再用Python生成loom文件，实在有点繁琐，那可不可以直接从Seurat 对象生成loom文件呢？答案当然是可以，使用SCopeLoomR即可。只需要使用build_loom函数，在get_count_from_seurat.R文件后面添加几行代码即可。
get_count_from_seurat.R文件内容如下：

library(optparse)
op_list <- list(
make_option(c("-i", "--inrds"), type = "character", default = NULL, action = "store", help = "The input of Seurat RDS",metavar="rds"),
make_option(c("-d", "--ident"), type = "character", default = NULL, action = "store", help = "The sample Ident of Seurat object",metavar="idents"),
make_option(c("-s", "--size"),  type = "integer", default = NULL, action = "store", help = "The sample size of Seurat object",metavar="size"),
make_option(c("-l", "--label"), type = "character", default = "out", action = "store", help = "The label of output file",metavar="label"),
make_option(c("-a", "--assay"), type = "character", default = "Spatial", action = "store", help = "The assay of input file",metavar="assay")
)
parser <- OptionParser(option_list = op_list)
opt = parse_args(parser)

assay <- opt$assay

library(Seurat)
obj <- readRDS(opt$inrds)
if (!is.null(opt$ident)) {
Idents(obj) <-  opt$ident
size=opt$size
if (!is.null(size)) {
obj <- subset(x = obj, downsample = opt$size)
}
saveRDS(obj,"subset.rds")
}
if (is.null(opt$label)) {
label1 <- 'out'
}else{
label1 <- opt$label
}

library(SCopeLoomR)
outloom <- paste0(label1,".loom")
build_loom(file.name = outloom,dgem = obj@assays[[assay]]@counts)
write.table([email protected],'metadata_subset.xls',sep='\t',quote=F)

使用方法：
-i，输入Seurat对象的RDS文件
-d，随机取样分组的列名，例如groups，如果不赋值则表示不随机取样，使用全部细胞
-s，随机取样的大小，例如20，因为这里用的是pbmc_small，所以设置小一点，实际情况可能需要设置大一点
-l，输出文件的标签，默认为out,
-a，使用的Seurat对象assay，默认为Spatial。

Rscript get_count_from_seurat.R -i test.rds -d groups -s 20 -l out -a Spatial

需要注意的是，新脚本添加了assay这个参数，默认为Spatial，因为最近在用跑空间组数据，如果要应用到单细胞需要改为RNA。

---

环境构建遇到的一些errors

因为最近集群在改系统盘名称还有不断删文件导致我用别人的软件、数据库文件和库路径时总是报错，因为自己开始来重新构建SCENIC/pyscenic运行需要的环境，期间遇到不少问题，这里汇总一下。

• 首先，需要安装一下miniconda，在这个网站下载根据网站教程进行安装
  https://docs.conda.io/en/latest/miniconda.html#linux-installers
  其实就是下载一个shell脚本，然后运行就好，例如：

sh Miniconda3-latest-Linux-x86_64.sh

创建一个环境，安装Python和R

conda create -n pyscenic
conda activate pyscenic
conda install r-base=4.2.1
conda install python=3.9.13
pip install pyscenic -i https://mirrors.aliyun.com/pypi/simple/
pip install pandas -i https://mirrors.aliyun.com/pypi/simple/
pip install scanpy -i https://mirrors.aliyun.com/pypi/simple/
pip install opencv-python -i https://mirrors.aliyun.com/pypi/simple/
pip install loompy -i https://mirrors.aliyun.com/pypi/simple/

这样就配置好基本环境了，接下来是安装R包，确保以下需要library的包就可以。

library(Seurat)
library(SCopeLoomR)
library(AUCell)
library(SCENIC)
library(dplyr)
library(KernSmooth)
library(RColorBrewer)
library(plotly)
library(BiocParallel)
library(optparse)

可以先用BiocManager::install() 安装，如果安装不了，谷歌搜一下conda install [package name]（百度搜不出来，不推荐）就能找到conda 安装这个包的官网，虽然conda很万能，但是慢啊，BiocManager::install会快很多。当然还有一些包是装不了，需要用devtools安装，例如SCopeLoomR，那哪些包要用devtools装呢？一般BiocManager::install和conda装不了的包就极大概率要用devtools装了，所以直接搜这个包的GitHub或官网，例如谷歌搜SCopeLoomR GitHub，GitHub里面说明需要这样安装：

devtools::install_github("aertslab/SCopeLoomR")

• 谈谈SCopeLoomR安装过程遇到的一些坑
  本来想偷懒直接用别人的库路径，library的时候结果报错HDF5 library version mismatched error，R还abort突然中断了，就是说我的HDF5库版本跟我这个包安装时的版本不匹配，只能重装。用devtools重装又各种报错，缺失某个包，缺失某个动态库……反正各种缺失，没办法，只能缺什么装什么了，而且装devtools的时候就很容易报错，建议先用conda安装以下包和动态库再安装SCopeLoomR：

conda install -c anaconda git
conda install -c anaconda hdf5
conda install -c conda-forge libgit2
conda install -c conda-forge r-ragg
conda install -c conda-forge r-xml

再在R里运行：

devtools::install_github("aertslab/SCopeLoomR")

---

SCENIC版本的bugs

在进行可视化的时候需要用到SCENIC的R包，但是我发现调函数plotRSS的过滤参数时，不管怎么调，出来的图都是一样的。因为看一下这个函数的源代码，好家伙，发现它莫名奇妙过滤了1.5的值，而且这个值是固定在函数里无法更改。

• 查看SCENIC版本1.2.4

> packageVersion("SCENIC")
[1] ‘1.2.4’

函数plotRSS源代码

> plotRSS
function (rss, labelsToDiscard = NULL, zThreshold = 1, cluster_columns = FALSE,
    order_rows = TRUE, trh = 0.01, varName = "cellType", col.low = "grey90",
    col.mid = "darkolivegreen3", col.high = "darkgreen", revCol = FALSE)
{
    varSize = "RSS"
    varCol = "Z"
    if (revCol) {
        varSize = "Z"
        varCol = "RSS"
    }
    rssNorm <- scale(rss)
    rssNorm[rssNorm < zThreshold] <- 0
    rssNorm <- rssNorm[, which(!colnames(rssNorm) %in% labelsToDiscard)]
    tmp <- .plotRSS_heatmap(rssNorm, trh = trh, cluster_columns = cluster_columns,
        order_rows = order_rows)
    rowOrder <- rev(tmp@row_names_param$labels)
    rss.df <- reshape2::melt(rss)
    head(rss.df)
    colnames(rss.df) <- c("Topic", varName, "RSS")
    rssNorm.df <- reshape2::melt(rssNorm)
    colnames(rssNorm.df) <- c("Topic", varName, "Z")
    rss.df <- merge(rss.df, rssNorm.df)
    rss.df <- rss.df[which(rss.df$Z >= 1.5), ]
    rss.df <- rss.df[which(!rss.df[, varName] %in% labelsToDiscard),
        ]
    rss.df[, "Topic"] <- factor(rss.df[, "Topic"], levels = rowOrder)
    p <- dotHeatmap(rss.df, var.x = varName, var.y = "Topic",
        var.size = varSize, min.size = 0.5, max.size = 5, var.col = varCol,
        col.low = col.low, col.mid = col.mid, col.high = col.high)
    invisible(list(plot = p, df = rss.df, rowOrder = rowOrder))
}

于是，我把这个包更新了一些再看函数代码，发现去掉了过滤1.5值的操作，果然是旧版本的一个bug啊。

• 查看SCENIC版本1.3.1

> packageVersion("SCENIC")
[1] ‘1.3.1’

函数plotRSS源代码

> plotRSS
function (rss, labelsToDiscard = NULL, zThreshold = 1, cluster_columns = FALSE,
    order_rows = TRUE, thr = 0.01, varName = "cellType", col.low = "grey90",
    col.mid = "darkolivegreen3", col.high = "darkgreen", revCol = FALSE,
    verbose = TRUE)
{
    varSize = "RSS"
    varCol = "Z"
    if (revCol) {
        varSize = "Z"
        varCol = "RSS"
    }
    rssNorm <- scale(rss)
    rssNorm <- rssNorm[, which(!colnames(rssNorm) %in% labelsToDiscard)]
    rssNorm[rssNorm < 0] <- 0
    rssSubset <- rssNorm
    if (!is.null(zThreshold))
        rssSubset[rssSubset < zThreshold] <- 0
    tmp <- .plotRSS_heatmap(rssSubset, thr = thr, cluster_columns = cluster_columns,
        order_rows = order_rows, verbose = verbose)
    rowOrder <- rev(tmp@row_names_param$labels)
    rm(tmp)
    rss.df <- reshape2::melt(rss)
    head(rss.df)
    colnames(rss.df) <- c("Topic", varName, "RSS")
    rssNorm.df <- reshape2::melt(rssNorm)
    colnames(rssNorm.df) <- c("Topic", varName, "Z")
    rss.df <- base::merge(rss.df, rssNorm.df)
    rss.df <- rss.df[which(!rss.df[, varName] %in% labelsToDiscard),
        ]
    if (nrow(rss.df) < 2)
        stop("Insufficient rows left to plot RSS.")
    rss.df <- rss.df[which(rss.df$Topic %in% rowOrder), ]
    rss.df[, "Topic"] <- factor(rss.df[, "Topic"], levels = rowOrder)
    p <- dotHeatmap(rss.df, var.x = varName, var.y = "Topic",
        var.size = varSize, min.size = 0.5, max.size = 5, var.col = varCol,
        col.low = col.low, col.mid = col.mid, col.high = col.high)
    invisible(list(plot = p, df = rss.df, rowOrder = rowOrder))
}

---

最后脚本汇总

三步完成单细胞数据调控网络流程分析和可视化

需要四个脚本文件：create_loom_input.R，pyscenic_from_loom.sh，calcRSS_by_scenic.R和function_pyscenic_visualize.R，我把四个脚本都放在目录_{/00.scripts/下，然后只需要输入Seurat对象的rds路径inrds（例如}/test.rds）和细胞分组列ingroup（例如sub），
运行示例：

source ~/Miniconda/md/bin/activate pyscenic

inscp='~/00.scripts/'
inrds='~/test.rds'
ingroup='sub'

#Step 1
Rscript ${inscp}/create_loom_input.R -i ${inrds} -d ${ingroup} -l out -a Spatial
#Step 2
sh ${inscp}/pyscenic_from_loom.sh -i out.loom -n 20
#Step 3
Rscript ${inscp}/calcRSS_by_scenic.R -l aucell.loom -m metadata_subset.xls -c ${ingroup}

运行上述脚本即可，有关参数请查看这篇博客四步完成单细胞数据调控网络流程分析-SCENIC/pySCENIC-2022-09-06，这里不再赘述。
请注意，这里的脚本是适用于小鼠的，如果要用人的请替换数据库三个文件。

---

附上需要用到的4个文件

• create_loom_input.R

library(optparse)
op_list <- list(
make_option(c("-i", "--inrds"), type = "character", default = NULL, action = "store", help = "The input of Seurat RDS",metavar="rds"),
make_option(c("-d", "--ident"), type = "character", default = NULL, action = "store", help = "The sample Ident of Seurat object",metavar="idents"),
make_option(c("-s", "--size"),  type = "integer", default = NULL, action = "store", help = "The sample size of Seurat object",metavar="size"),
make_option(c("-l", "--label"), type = "character", default = "out", action = "store", help = "The label of output file",metavar="label"),
make_option(c("-a", "--assay"), type = "character", default = "Spatial", action = "store", help = "The assay of input file",metavar="assay")
)
parser <- OptionParser(option_list = op_list)
opt = parse_args(parser)

assay <- opt$assay

library(Seurat)
obj <- readRDS(opt$inrds)
if (!is.null(opt$ident)) {
Idents(obj) <-  opt$ident
size=opt$size
if (!is.null(size)) {
obj <- subset(x = obj, downsample = opt$size)
}
saveRDS(obj,"subset.rds")
}
if (is.null(opt$label)) {
label1 <- 'out'
}else{
label1 <- opt$label
}

library(SCopeLoomR)
outloom <- paste0(label1,".loom")
build_loom(file.name = outloom,dgem = obj@assays[[assay]]@counts)
write.table([email protected],'metadata_subset.xls',sep='\t',quote=F)

• pyscenic_from_loom.sh

input_loom=out.loom
n_workers=20
#help function
function usage() {
echo -e "OPTIONS:\n-i|--input_loom:\t input loom file"
echo -e "-n|--n_workers:\t working core number"
echo -e "-h|--help:\t Usage information"
exit 1
}
#get value
while getopts :i:n:h opt
do
    case "$opt" in
        i) input_loom="$OPTARG" ;;
        n) n_workers="$OPTARG" ;;
        h) usage ;;
        :) echo "This option -$OPTARG requires an argument."
           exit 1 ;;
        ?) echo "-$OPTARG is not an option"
           exit 2 ;;
    esac
done
#在这里改数据库文件
database='~/01.mouse/'
tfs=${database}/allTFs_mm.txt
feather=${database}/mm10_10kbp_up_10kbp_down_full_tx_v10_clust.genes_vs_motifs.rankings.feather
tbl=${database}/motifs-v10nr_clust-nr.mgi-m0.001-o0.0.tbl.txt
pyscenic=pyscenic

# grn
$pyscenic grn \
--num_workers $n_workers \
--output grn.tsv \
--method grnboost2 \
$input_loom  $tfs

# cistarget
$pyscenic ctx \
grn.tsv $feather \
--annotations_fname $tbl \
--expression_mtx_fname $input_loom \
--mode "dask_multiprocessing" \
--output ctx.csv \
--num_workers $n_workers   \
--mask_dropouts

# AUCell
$pyscenic aucell \
$input_loom \
ctx.csv \
--output aucell.loom \
--num_workers $n_workers

• calcRSS_by_scenic.R

library(optparse)
op_list <- list(
make_option(c("-l", "--input_loom"), type = "character", default = NULL, action = "store", help = "The input of aucell loom file",metavar="rds"),
make_option(c("-m", "--input_meta"), type = "character", default = NULL, action = "store", help = "The metadata of Seurat object",metavar="idents"),
make_option(c("-a", "--assay"), type = "character", default = 'Spatial', action = "store", help = "The assay of Seurat object",metavar="assay"),
make_option(c("-c", "--celltype"), type = "character", default = NULL, action = "store", help = "The colname of metadata to calculate RSS",metavar="label")
)
parser <- OptionParser(option_list = op_list)
opt = parse_args(parser)

library(Seurat)
library(SCopeLoomR)
library(AUCell)
library(SCENIC)
library(dplyr)
library(KernSmooth)
library(RColorBrewer)
library(plotly)
library(BiocParallel)

celltype <- opt$celltype
assay <- opt$assay
loom <- open_loom(opt$input_loom)

regulons_incidMat <- get_regulons(loom, column.attr.name="Regulons")
regulons <- regulonsToGeneLists(regulons_incidMat)
regulonAUC <- get_regulons_AUC(loom,column.attr.name='RegulonsAUC')
regulonAucThresholds <- get_regulon_thresholds(loom)
close_loom(loom)

meta <- read.table(opt$input_meta,sep='\t',header=T,stringsAsFactor=F)
cellinfo <- meta[,c(opt$celltype,paste0("nFeature_",assay),paste0("nCount_",assay))]
colnames(cellinfo)=c('celltype', 'nGene' ,'nUMI')
cellTypes <-  as.data.frame(subset(cellinfo,select = 'celltype'))
selectedResolution <- "celltype"

sub_regulonAUC <- regulonAUC
rss <- calcRSS(AUC=getAUC(sub_regulonAUC),
               cellAnnotation=cellTypes[colnames(sub_regulonAUC),
                                        selectedResolution])
rss=na.omit(rss)
try({
rssPlot <- plotRSS(rss)
save(regulonAUC,rssPlot,regulons,file='regulon_RSS.Rdata')
})

saveRDS(rss,paste0(celltype,"_rss.rds"))

source('/hwfssz1/ST_SUPERCELLS/P22Z10200N0664/huangbaoqian/ST_PRECISION_USER_huangbaoqian/SCENIC/04.database/00.common_scripts/function_pyscenic_visualize.R')
plot_pyscenic(inloom='aucell.loom',incolor=incolor,inrss=paste0(celltype,"_rss.rds"),inrds='subset.rds',infun='median', ct.col=celltype,inregulons=NULL,ingrn='grn.tsv',ntop1=5,ntop2=50)

• function_pyscenic_visualize.R

library(Seurat)
library(SCopeLoomR)
library(AUCell)
library(SCENIC)
library(dplyr)
library(KernSmooth)
library(RColorBrewer)
library(plotly)
library(BiocParallel)
library(pheatmap)

library(cowplot)
library(ggpubr)
library(ggsci)
library(ggplot2)
library(tidygraph)
library(ggraph)
library(stringr)



colpalettes<-unique(c(pal_npg("nrc")(10),pal_aaas("default")(10),pal_nejm("default")(8),pal_lancet("lanonc")(9),
                      pal_jama("default")(7),pal_jco("default")(10),pal_ucscgb("default")(26),pal_d3("category10")(10),
                      pal_locuszoom("default")(7),pal_igv("default")(51),
                      pal_uchicago("default")(9),pal_startrek("uniform")(7),
                      pal_tron("legacy")(7),pal_futurama("planetexpress")(12),pal_rickandmorty("schwifty")(12),
                      pal_simpsons("springfield")(16),pal_gsea("default")(12)))
len <- 100
incolor<-c(brewer.pal(8, "Dark2"),brewer.pal(12, "Paired"),brewer.pal(8, "Set2"),brewer.pal(9, "Set1"),colpalettes,rainbow(len))

plot_pyscenic <- function(inloom='aucell.loom',incolor=incolor,inrss='seurat_annotations_rss.rds',inrds='subset.rds',infun='median', ct.col='seurat_annotations',inregulons=NULL,ingrn='grn.tsv',ntop1=5,ntop2=50){
###load data
loom <- open_loom(inloom)

regulons_incidMat <- get_regulons(loom, column.attr.name="Regulons")
regulons <- regulonsToGeneLists(regulons_incidMat)
regulonAUC <- get_regulons_AUC(loom,column.attr.name='RegulonsAUC')
regulonAucThresholds <- get_regulon_thresholds(loom)

embeddings <- get_embeddings(loom)
close_loom(loom)

rss <- readRDS(inrss)
sce <- readRDS(inrds)

##calculate  RSS fc
df = do.call(rbind,
             lapply(1:ncol(rss), function(i){
                dat= data.frame(
                regulon  = rownames(rss),
                cluster =  colnames(rss)[i],
                sd.1 = rss[,i],
                sd.2 = apply(rss[,-i], 1, get(infun))
               )
             }))

df$fc = df$sd.1 - df$sd.2

#select top regulon
ntopg <- df %>% group_by(cluster) %>% top_n(ntop1, fc)

ntopgene <- unique(ntopg$regulon)
write.table(ntopgene,'sd_regulon_RSS.list',sep='\t',quote=F,row.names=F,col.names=F)
#plot rss by cluster

#using plotRSS
rssPlot <- plotRSS(rss)
regulonsToPlot <- rssPlot$rowOrder
rp_df <- rssPlot$df

write.table(regulonsToPlot,'rss_regulon.list',sep='\t',quote=F,row.names=F,col.names=F)
write.table(rp_df,'rssPlot_data.xls',sep='\t',quote=F)
nlen <- length(regulonsToPlot)
hei <- ceiling(nlen)*0.4
blu<-colorRampPalette(brewer.pal(9,"Blues"))(100)
lgroup <- levels(rssPlot$df$cellType)

nlen2 <- length(lgroup)
wei <- nlen2*2
pdf(paste0('regulons_RSS_',ct.col,'_in_dotplot.pdf'),wei,hei)
print(rssPlot$plot)
dev.off()

# sd top gene
anrow = data.frame( group = ntopg$cluster)
lcolor <- incolor[1:length(unique(ntopg$cluster))]
names(lcolor) <- unique(anrow$group)
annotation_colors <- list(group=lcolor)

pn1 = rss[ntopg$regulon,]
pn2 = rss[unique(ntopg$regulon),]
rownames(pn1) <-  make.unique(rownames(pn1))
rownames(anrow) <- rownames(pn1)
scale='row'
hei <- ceiling(length(ntopg$regulon)*0.4)
pdf(paste0('regulon_RSS_in_sd_topgene_',ct.col,'.pdf'),wei,hei)
print(
pheatmap(pn1,annotation_row = anrow,scale=scale,annotation_colors=annotation_colors,show_rownames = T,main='sd top regulons')
)
print(
pheatmap(pn2,scale=scale,show_rownames = T, main='sd top unique regulons')
)
dev.off()

#plotRSS gene

pn2 = rss[unique(rp_df$Topic),]
scale='row'
hei <- ceiling(length(unique(rp_df$Topic))*0.4)
pdf(paste0('regulon_RSS_in_plotRSS_',ct.col,'.pdf'),wei,hei)
print(
pheatmap(pn2,scale=scale,show_rownames = T, main='plotRSS unique regulons')
)
dev.off()

#all regulons

hei <- ceiling(length(rownames(rss))*0.2)
pdf(paste0('all_regulons_RSS_in_',ct.col,'.pdf'),wei,hei)
print(
pheatmap(rss,scale=scale,show_rownames = T,main='all regulons RSS')
)
dev.off()
#plot rss by all cells
if (is.null(inregulons)){
inregulons <- regulonsToPlot
}else{
inregulons <- intersect(inregulons,rownames(rss))
regulonsToPlot <- inregulons

}
pn3=as.matrix(regulonAUC@assays@data$AUC)
regulon <- rownames(pn3)
#regulon <- inregulons
pn3 <- pn3[regulon,]
#pn3 <- pn3[,sample(1:dim(pn3)[2],500)]

[email protected][,ct.col]

meta <- [email protected]
meta <- meta[order(meta$group1),]
#meta <- meta[colnames(pn3),]
ancol = data.frame(meta[,c('group1')])
colnames(ancol) <- c('group1')
rownames(ancol) <- rownames(meta)
lcolor <- incolor[1:length(unique(ntopg$cluster))]
names(lcolor) <- unique(ntopg$cluster)
annotation_colors <- list(group1 =lcolor)

df1 <- ancol
df1$cell <- rownames(df1)
df1 <- df1[order(df1$group1),]
pn3 <- pn3[,rownames(df1)]
torange=c(-2,2)
pn3 <- scales::rescale(pn3,to=torange)
pn3 <- pn3[,rownames(ancol)]

scale='none'
hei <- ceiling(length(unique(regulon))*0.2)
pdf(paste0('all_regulon_activity_in_allcells.pdf'),10,hei)
print(
pheatmap(pn3,annotation_col = ancol,scale=scale,annotation_colors=annotation_colors,show_rownames = T,show_colnames = F,cluster_cols=F)
)
#pheatmap(pn3,scale=scale,show_rownames = T, show_colnames = F,cluster_cols=F)
dev.off()

#plot in seurat
regulonsToPlot = inregulons
sce$sub_celltype <- [email protected][,ct.col]
sub_regulonAUC <- regulonAUC[,match(colnames(sce),colnames(regulonAUC))]

cellClusters <- data.frame(row.names = colnames(sce),
                           seurat_clusters = as.character(sce$seurat_clusters))
cellTypes <- data.frame(row.names = colnames(sce),
                        celltype = sce$sub_celltype)

[email protected] = cbind([email protected] ,t(sub_regulonAUC@assays@data@listData$AUC[regulonsToPlot,]))
Idents(sce) <- sce$sub_celltype

nlen <- length(regulonsToPlot)
hei <- ceiling(nlen)*0.4
blu<-colorRampPalette(brewer.pal(9,"Blues"))(100)
nlen2 <- length(unique(sce$sub_celltype))
wei <- nlen2*2
pdf('regulons_activity_in_dotplot.pdf',wei,hei)
print(DotPlot(sce, features = unique(regulonsToPlot)) + coord_flip()+
      theme(axis.text.x = element_text(angle = 90, hjust = 1, vjust = .5))+
      scale_color_gradientn(colours = blu)
      )
dev.off()

hei=ceiling(nlen/4)*4
pdf('regulons_activity_in_umap.pdf',16,hei)
print(RidgePlot(sce, features = regulonsToPlot , ncol = 4))
print(VlnPlot(sce, features = regulonsToPlot,pt.size = 0 ))
print(FeaturePlot(sce, features = regulonsToPlot))
dev.off()

grn <- read.table(ingrn,sep='\t',header=T,stringsAsFactors=F)
inregulons1=gsub('[(+)]','',inregulons)

c1 <- which(grn$TF %in% inregulons1)
grn <- grn[c1,]
#edge1 <- data.frame()
#node1 <- data.frame()
pdf(paste0(ntop2,'_regulon_netplot.pdf'),10,10)
for (tf in unique(grn$TF)) {
tmp <- subset(grn,TF==tf)
if (dim(tmp)[1] > ntop2) {
tmp <- tmp[order(tmp$importance,decreasing=T),]
tmp <- tmp[1:ntop2,]
}
node2 <- data.frame(tmp$target)
node2$node.size=1.5
node2$node.colour <- 'black'
colnames(node2) <- c('node','node.size','node.colour')
df1 <- data.frame(node=tf,node.size=2,node.colour='#FFDA00')
node2 <- rbind(df1,node2)


edge2 <- tmp
colnames(edge2) <- c('from','to','edge.width')
edge2$edge.colour <- "#1B9E77"
torange=c(0.1,1)
edge2$edge.width <- scales::rescale(edge2$edge.width,to=torange)

graph_data <- tidygraph::tbl_graph(nodes = node2, edges = edge2, directed = T)
p1 <- ggraph(graph = graph_data, layout = "stress", circular = TRUE) + geom_edge_arc(aes(edge_colour = edge.colour, edge_width = edge.width)) +
  scale_edge_width_continuous(range = c(1,0.2)) +geom_node_point(aes(colour = node.colour, size = node.size))+ theme_void() +
      geom_node_label(aes(label = node,colour = node.colour),size = 3.5, repel = TRUE)
p1 <- p1 + scale_color_manual(values=c('#FFDA00','black'))+scale_edge_color_manual(values=c("#1B9E77"))
print(p1)
}
dev.off()
#plot activity heatmap
meta <- [email protected]
celltype <- ct.col
cellsPerGroup <- split(rownames(meta),meta[,celltype])
sub_regulonAUC <- regulonAUC[onlyNonDuplicatedExtended(rownames(regulonAUC)),]
# Calculate average expression:
regulonActivity_byGroup <- sapply(cellsPerGroup,
                                  function(cells)
                                    rowMeans(getAUC(sub_regulonAUC)[,cells]))
scale='row'
rss <- regulonActivity_byGroup
hei <- ceiling(length(regulonsToPlot)*0.4)
pn1 <- rss[regulonsToPlot,]
pdf(paste0('regulon_activity_in_',ct.col,'.pdf'),wei,hei)
print(
pheatmap(pn1,scale=scale,show_rownames = T, main='regulons activity')
)
dev.off()

hei <- ceiling(length(rownames(rss))*0.2)
pdf(paste0('all_regulons_activity_in_',ct.col,'.pdf'),wei,hei)
print(
pheatmap(rss,scale=scale,show_rownames = T,main='all regulons activity')
)
dev.off()

##calculate fc
df = do.call(rbind,
             lapply(1:ncol(rss), function(i){
                dat= data.frame(
                regulon  = rownames(rss),
                cluster =  colnames(rss)[i],
                sd.1 = rss[,i],
                sd.2 = apply(rss[,-i], 1, get(infun))
               )
             }))

df$fc = df$sd.1 - df$sd.2

#select top regulon
ntopg <- df %>% group_by(cluster) %>% top_n(ntop1, fc)

ntopgene <- unique(ntopg$regulon)
write.table(ntopgene,'sd_regulon_activity.list',sep='\t',quote=F,row.names=F,col.names=F)

anrow = data.frame( group = ntopg$cluster)
lcolor <- incolor[1:length(unique(ntopg$cluster))]
names(lcolor) <- unique(anrow$group)
annotation_colors <- list(group=lcolor)
pn1 = rss[ntopg$regulon,]
pn2 = rss[unique(ntopg$regulon),]
rownames(pn1) <-  make.unique(rownames(pn1))
rownames(anrow) <- rownames(pn1)
scale='row'
hei <- ceiling(length(ntopg$regulon)*0.4)
pdf(paste0('regulon_activity_in_sd_topgene_',ct.col,'.pdf'),wei,hei)
print(
pheatmap(pn1,annotation_row = anrow,scale=scale,annotation_colors=annotation_colors,show_rownames = T,main='sd top regulons')
)
print(
pheatmap(pn2,scale=scale,show_rownames = T, main='sd top unique regulons ')
)
dev.off()

}