章峰的网站 - 生信

Visium空间转录组分析流程

2023-05-09T00:00:00+08:00

# Analysis, visualization, and integration of spatial datasets with Seurat at https://satijalab.org/seurat/articles/spatial_vignette.html
# set path ------------------------------------------
rm(list=ls());options(stringsAsFactors=FALSE)
project_dir <- rprojroot::find_rstudio_root_file()
raw_dir <- paste0(project_dir,"/raw-data/")
process_dir <- paste0(project_dir,"/process-data/")
output_dir <- paste0(project_dir,"/output-results/")
temp_dir <- paste0(project_dir,"/temp-data/")
bin_dir <- paste0(project_dir,"/bin/")
script_dir <- paste0(project_dir,"/workflow-scripts/")
work_dir <- output_dir
setwd(work_dir)

# library and set default parameter  ----------------
library(tidyverse)
library(Seurat)
library(SeuratData) #devtools::install_github('satijalab/seurat-data')
library(patchwork)
library(edgeR)
library(SingleCellExperiment)
prefix <- "Seurat-3-"
output_prefix <- paste0(output_dir,prefix)

创建一些函数¶

# functions -------------
run_clustering <- function(data, resolution = 0.1, npcs=30) {

  data <- SCTransform(data, assay = "Spatial", verbose = TRUE)
  data <- RunPCA(data, assay = "SCT", verbose = TRUE)
  data <- FindNeighbors(data, reduction = "pca", dims = 1:npcs)
  data <- FindClusters(data, verbose = TRUE, resolution=resolution)
  data <- RunUMAP(data, dims = 1:npcs)
  return(data)
}

pooling_seu <- function(seu_obj,assay="SCT") {
  cell_metadata <- seu_obj@meta.data

  # create sudo id by random sample into 3 sets from each patients
  cell_metadata$sudo_sample_id <- NA
  samples <- unique(cell_metadata$orig.ident)
  nrg <- 3
  for(i in c(1:length(samples))){
    set.seed(i)
    temp_sample_idx <- which(cell_metadata$orig.ident==samples[i])
    sample_sudo_idx <- sample(c(1:nrg),length(temp_sample_idx),replace=T,prob=rep(1/nrg,nrg)) # random sample into 3 sets
    cell_metadata[temp_sample_idx,'sudo_sample_id'] <- paste(samples[i],sample_sudo_idx,sep='_')
  }
  sce <- SingleCellExperiment(assays = list(counts = GetAssayData(seu_obj,slot="count",assay=assay)), colData = cell_metadata)
  summed <- scuttle::aggregateAcrossCells(sce, id=colData(sce)[,'sudo_sample_id']) # aggregate counts by id
  return(summed)
}

dga_treat <- function(DGEList,ref_level=NA,set_regulate=TRUE){
  DGEList$samples$lib.size <- colSums(DGEList$counts)
  if(is.na(ref_level)){
    DGEList$samples$group <- relevel(DGEList$samples$group, ref = levels(DGEList$samples$group)[1] ) 
  } else{
    DGEList$samples$group <- relevel(DGEList$samples$group, ref = ref_level)
  }
  keep <- filterByExpr(DGEList, group=DGEList$samples$group)   #set your own threshold
  table(keep)
  DGEList <- DGEList[keep,, keep.lib.sizes=FALSE]
  DGEList <- calcNormFactors(DGEList)
  # DE
  design <- model.matrix(~DGEList$samples$group)
  DGEList <- estimateDisp(DGEList, design, robust=TRUE)
  fit <- glmQLFit(DGEList, design, robust=TRUE)
  #lrt <- glmQLFTest(fit, coef=ncol(fit$design))
  lrt <- glmTreat(fit, coef=ncol(fit$design), lfc=log2(1.2))
  print(table(decideTests(lrt)))
  res <- topTags(lrt,n = nrow(DGEList))
  if(set_regulate){
    res$table$regulate <- "Normal"
    res$table$regulate[res$table$logFC>0 & res$table$FDR<0.05] <- "Up"
    res$table$regulate[res$table$logFC<0 & res$table$FDR<0.05] <- "Down"
    res$table <- res$table[order(res$table$FDR),]
  }
  return(res)
}

plot_rctd <- function(RCTD,seu_obj, image_name="VLP43_kidney_A1",r=3){
  stopifnot("RCTD"%in% class(RCTD))
  stopifnot("FindClusters"%in% names(seu_obj@commands))
  norm_weights <- normalize_weights(RCTD@results$weights) # normalize the cell type proportions to sum to 1.
  query <- subset(seu_obj,cells = row.names(RCTD@spatialRNA@coords))
  pos <- RCTD@spatialRNA@coords
  annot <- query$seurat_clusters
  p <- vizAllTopics(as.matrix(norm_weights), pos,
                    #topicCols = qualitative_hcl(n=ncol(norm_weights),palette="Dynamic"),
                    topicCols = brewer.pal(n=ncol(norm_weights),name = "Paired"),
                    groups = annot,
                    group_cols = rainbow(length(levels(annot))),
                    #group_cols = rainbow_hcl(length(levels(annot))),
                    r=r,lwd = 0.1) + labs(title=image_name) +
    #ggplot2::guides(colour = "none") +
    theme(plot.background = ggplot2::element_rect(fill = "white"),plot.title = element_text(hjust = 0.5,size=15))
  return(p)
}

Visium空间转录组技术¶

当前时代的空间转录组技术大致分为五类方向：激光显微切割（laser capture microdissection，LCM），单分子荧光原位杂交（single molecular fluorescent in situ hybridization, smFISH），靶向原位测序（In situ sequencing，ISS），原位阵列捕获（In situ array capture, Array）和其他非成像技术（No imaging）。

目前商业用的最广的还是Visium平台，因此我们以Visium数据为例进行演示，进行如聚类，差异表达，反卷积等分析。使用的工具是Seurat，spacexr等包。在单细胞下游分析中，Seurat是一个非常强大的R包，不仅有强大的社区、详细文档说明还有很多内置数据。这里以Analysis, visualization, and integration of spatial datasets with Seurat 数据为例进行空间转录组分析。

1. 数据探索¶

查看测序数据质量，是否符合生物学背景。如果质量不好，需要进行质量控制。stxBrain已经是过滤后的数据，我们不需要进行质量控制。

#InstallData("stxBrain") # 如果由于网络问题，安装失败。可以将url放在浏览器中下载tar.gz文件，然后本地安装。
brain <- LoadData("stxBrain", type = "anterior1")

# Data preprocessing: Normalization 
plot1 <- VlnPlot(brain, features = "nCount_Spatial", pt.size = 0.1) + NoLegend()
plot2 <- SpatialFeaturePlot(brain, features = "nCount_Spatial") + theme(legend.position = "right")
wrap_plots(plot1, plot2)
# 斑点上分子计数的变化不仅是技术上的问题，而且还取决于组织的解剖结构。
# 例如，神经元（例如皮质白质）耗竭的组织区域可再现地显示出较低的分子数。

2. 聚类¶

和单细胞数据类似，我们需要对数据进行聚类。也可以使用run_clustering函数。

brain <- SCTransform(brain, assay = "Spatial", verbose = FALSE)
# Gene expression visualization
SpatialFeaturePlot(brain, features = c("Hpca", "Ttr"))
p1 <- SpatialFeaturePlot(brain, features = "Ttr", pt.size.factor = 1)
p2 <- SpatialFeaturePlot(brain, features = "Ttr", alpha = c(0.1, 1))
p1 + p2

# Dimensionality reduction, clustering 
brain <- RunPCA(brain, assay = "SCT", verbose = FALSE)
brain <- FindNeighbors(brain, reduction = "pca", dims = 1:30)
brain <- FindClusters(brain, verbose = FALSE)
brain <- RunUMAP(brain, reduction = "pca", dims = 1:30)

# visualization
p1 <- DimPlot(brain, reduction = "umap", label = TRUE)
p2 <- SpatialDimPlot(brain, label = TRUE, label.size = 3)
p1 + p2

3. 检测亚群特异和空间特异差异基因¶

对不同亚群进行差异分析，对不同空间分布中的细胞进行基因差异分析。

# Detecting spatially-variable features 
de_markers <- FindMarkers(brain, ident.1 = 5, ident.2 = 6)
head(de_markers)
SpatialFeaturePlot(object = brain, features = rownames(de_markers)[1:3], alpha = c(0.1, 1), ncol = 3)

A data.frame: 6 * 5
	p_val	avg_log2FC	pct.1	pct.2	p_val_adj

Calb2	6.427214e-69	3.336874	1.000	0.537	1.135560e-64
Camk2n1	1.519204e-68	-2.450388	1.000	1.000	2.684130e-64
Nrgn	1.573095e-68	-3.229826	0.971	1.000	2.779344e-64
Stx1a	2.104119e-68	-2.238353	0.784	1.000	3.717558e-64
Nptxr	9.820482e-68	-1.918293	0.942	1.000	1.735083e-63
Lingo1	2.124777e-67	-1.913281	0.880	1.000	3.754056e-63

#在没有预注释的情况下搜索表现出空间模式的特征
brain <- FindSpatiallyVariableFeatures(brain, assay = "SCT", features = VariableFeatures(brain)[1:1000],selection.method = "moransi") # moransi mothed, markvariogram too slow
#top.features <- head(SpatiallyVariableFeatures(brain, selection.method = "moransi"), 6)
top.features <- c("Calb2","Gng4","Ttr","S100a5","Nrgn","Doc2g")
SpatialFeaturePlot(brain, features = top.features, ncol = 3, alpha = c(0.1, 1))

Computing Moran's I

4. 检测分组特异差异基因¶

由于此数据没有分组信息，我们导入另一个数据进行演示。由于直接对分组细胞进行差异分析会造成假阳性，这里我们用psudo-bulk差异分析

brain <- readRDS("/media/zhangfeng/myData/projects/sp-aging/output-results/70-visium-brain-cluster.RData")

# DEG group--------
table(brain@meta.data$group)
summed <- pooling_seu(brain)
class(summed)
head(summed@colData)
is_ribo<-grepl('Rps|Rpl',row.names(summed))
is_mito<-grepl('mt-',row.names(summed))
y <- DGEList(counts(summed)[(!is_ribo)&(!is_mito),], samples=colData(summed)$sudo_sample_id,group=colData(summed)$group) # DE between types
res <- dga_treat(DGEList=y,ref_level = "young")
head(res$table)

aged young
10715 11479

DataFrame with 6 rows and 11 columns
            orig.ident nCount_Spatial nFeature_Spatial       group nCount_SCT`<character>`      `<numeric>`        `<integer>` `<character>`  `<numeric>`
VLP40_1A_1    VLP40_1A             NA               NA       young         NA
VLP40_1A_2    VLP40_1A             NA               NA       young         NA
VLP40_1A_3    VLP40_1A             NA               NA       young         NA
VLP40_1B_1    VLP40_1B             NA               NA       young         NA
VLP40_1B_2    VLP40_1B             NA               NA       young         NA
VLP40_1B_3    VLP40_1B             NA               NA       young         NA
           nFeature_SCT SCT_snn_res.0.1 seurat_clusters sudo_sample_id
              `<integer>`        `<factor>`        `<factor>`    `<character>`
VLP40_1A_1           NA              NA              NA     VLP40_1A_1
VLP40_1A_2           NA              NA              NA     VLP40_1A_2
VLP40_1A_3           NA              NA              NA     VLP40_1A_3
VLP40_1B_1           NA              NA              NA     VLP40_1B_1
VLP40_1B_2           NA              NA              NA     VLP40_1B_2
VLP40_1B_3           NA              NA              NA     VLP40_1B_3
                   ids    ncells
           `<character>` `<integer>`
VLP40_1A_1  VLP40_1A_1      1023
VLP40_1A_2  VLP40_1A_2      1041
VLP40_1A_3  VLP40_1A_3      1025
VLP40_1B_1  VLP40_1B_1       915
VLP40_1B_2  VLP40_1B_2       915
VLP40_1B_3  VLP40_1B_3       865

-1     0     1 
 87 14105   203

A data.frame: 6 * 6
	logFC	unshrunk.logFC	logCPM	PValue	FDR	regulate

Lyz2	2.6841911	2.6854073	5.326010	2.446786e-17	2.992338e-13	Up
C4b	3.0082676	3.0088154	6.819937	5.386843e-17	2.992338e-13	Up
Itgax	3.8597675	3.9129263	1.172032	6.236203e-17	2.992338e-13	Up
Cd52	1.3891682	1.3899204	4.541512	1.229941e-15	4.426248e-12	Up
Pcdhb9	0.9295853	0.9301349	4.305468	2.580643e-15	7.429671e-12	Up
Fcgr2b	1.4622417	1.4650595	2.751271	1.811064e-14	4.032808e-11	Up

5. Deconvolution分析¶

由于Visium平台的spot不是单细胞分辨率，一般有1-10个细胞，因此我们需要对其进行注释，分析不同细胞的比例。这里使用spacexr包进行分析。

reference <- readRDS('/media/zhangfeng/myData/projects/sp-aging/output-results/4-brain-SCRef.rds')
# control sample
spatial_controls <- RCTD_replicates(seu_obj = brain[,brain$group=="young"])
RCTD_controls <- create.RCTD.replicates(spatial_controls$spatialRNA_replicates, reference, spatial_controls$samples, max_cores = 20)
RCTD_controls <- run.RCTD.replicates(RCTD_controls, doublet_mode = 'full')

# treatment samples
spatial_treats <- RCTD_replicates(seu_obj = brain[,brain$group=="aged"])
RCTD_treats <- create.RCTD.replicates(spatial_treats$spatialRNA_replicates, reference, spatial_treats$samples, max_cores = 20)
RCTD_treats <- run.RCTD.replicates(RCTD_treats, doublet_mode = 'full')

# add into seurat
spot_weight <- combine_weight(RCTD_controls,RCTD_treats)
meta_data <- brain@meta.data %>% rownames_to_column("spot_id") %>% 
  left_join(spot_weight,by="spot_id")
all(meta_data$spot_id==colnames(brain))  
meta_data <- column_to_rownames(meta_data,"spot_id")
brain <- AddMetaData(brain,metadata = meta_data)
saveRDS(brain,file = "/media/zhangfeng/myData/projects/sp-aging/output-results/70-visium-brain-RCTD.RData"))

brain <- readRDS("/media/zhangfeng/myData/projects/sp-aging/output-results/70-visium-brain-RCTD.RData")

head(brain@meta.data)

A data.frame: 6 * 17
	orig.ident	nCount_Spatial	nFeature_Spatial	group	nCount_SCT	nFeature_SCT	SCT_snn_res.0.1	seurat_clusters	Astrocytes	Ependymal	Immune	Microglia	Neurons	Oligos	PeripheralGlia	Vascular	max_cell

AAACAAGTATCTCCCA-1_1_1_1	VLP40_1A	8421	3160	young	15335	3514	3	3	0.19143611	0.0006158916	0.0049827905	0.004197530	0.3236210	0.433052200	0.011450442	0.030644049	Oligos
AAACACCAATAACTGC-1_1_1_1	VLP40_1A	23006	5649	young	16862	5507	0	0	0.11293516	0.0013890113	0.0006617161	0.011798306	0.8054124	0.062050025	0.002048477	0.003704872	Neurons
AAACAGAGCGACTCCT-1_1_1_1	VLP40_1A	25131	6224	young	16782	5926	0	0	0.04871926	0.0073010556	0.0026329875	0.009357506	0.9132606	0.004827619	0.002313284	0.011587694	Neurons
AAACAGCTTTCAGAAG-1_1_1_1	VLP40_1A	35993	7074	young	16298	5253	0	0	0.05981760	0.0019247194	0.0001026660	0.013729242	0.8918254	0.007285013	0.002966418	0.022348983	Neurons
AAACAGGGTCTATATT-1_1_1_1	VLP40_1A	37738	7220	young	16163	5196	0	0	0.05777206	0.0014648800	0.0009153018	0.007876570	0.9005069	0.023966030	0.000675534	0.006822758	Neurons
AAACATTTCCCGGATT-1_1_1_1	VLP40_1A	14901	4544	young	15460	4544	3	3	0.13274107	0.0026031003	0.0001558520	0.006728277	0.6066684	0.218612193	0.010717393	0.021773681	Neurons

6. 绘图¶

slice_id <- names(RCTD_controls@group_ids)[1]
myRCTD_full <- RCTD_controls@RCTD.reps[[1]]
p_brain <- plot_rctd(myRCTD_full,brain,image_name = slice_id)+ggtitle(slice_id)+
    theme(plot.title=element_text(hjust = 0.5))+labs(fill="Cell_Type",color="Cluster")+
    guides(colour = "none")
p_brain

Anaconda使用教程

2021-10-21T10:20:00+08:00

[toc]

Anaconda使用教程¶

参考资料¶

中文博客推荐 anaconda 官网教程

一、什么是Anaconda？¶

1. 简介¶

Anaconda（官方网站）就是可以便捷获取包且对包能够进行管理，同时对环境可以统一管理的发行版本。Anaconda包含了conda、Python在内的超过180个科学包及其依赖项。

Anaconda可以创建不同的虚拟环境，以实现环境间的隔离，这样在不同环境中可以使用软件或依赖库的不同版本，避免不同版本间的冲突。

2. 特点¶

Anaconda具有如下特点：

开源
安装过程简单
高性能使用Python和R语言
免费的社区支持
适用于多平台：Windows, macOS, Linux

如果日常工作或学习并不必要使用1,000多个库，那么可以考虑安装Miniconda，这里不过多介绍Miniconda的安装及使用。

二、安装及镜像¶

1. 下载Anaconda¶

Anaconda官网下载地址： https://www.anaconda.com/products/individual#linux Anaconda中国镜像下载： https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/ miniconda下载地址： https://docs.conda.io/en/latest/miniconda.html#latest-miniconda-installer-links

2. 安装Anaconda¶

bash Anaconda3-2021.05-Linux-x86_64.sh  
# specify a different location below  
>>> /home/zhangfeng/bin/anaconda3

注意指定安装目录时，不能进行修改。如果有错字，使用删除键也不行。

3. 设置镜像¶

显示镜像：conda config --show-source

默认有以下载通道：

channels:
  - conda-forge
  - bioconda
  - r
  - defaults
show_channel_urls: true

即从国外的服务器下载不同软件包。- r表示通过r通道下载软件，即指从 https://conda.anaconda.org/r/网站下载 conda-forge是指从 https://conda.anaconda.org/conda-forge/网站下载

默认下载网址为Anaconda的官方服务器，安装包时下载速度很慢，所以设置国内镜像很重要。清华大学的TUNA镜像源有Anaconda仓库的镜像，我们将其加入conda的配置即可，即修改~/.condarc文件内容。conda会从.condarc中下载通道按顺序搜索软件包。 defaults ：默认下载通道。如果不指定通道则从defaults通道下载。 channel_alias：修改指向defaults的网址，我们可以用 https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/镜像代替。

修改后的.condarc文件：

channel_alias: https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
channels:
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/r
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/pro
  - https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/msys2
  - conda-forge
  - bioconda
  - defaults
ssl_verify: true
show_channel_urls: true
channel_priority: disabled

conda-forge和bioconda通道里的软件包比较新，推荐使用。

4. 安装软件包注意事项¶

安装一个包不成功的话，可能的原因有：

在本平台（windows）上没有这个包，或没有相应的硬件支持。
有依赖的包没有安装，可以网上查询一下可能的依赖包，先安装这个依赖包。
如果之前安装好的，但由于其它软件的变动，可能导致无法使用。可以先卸载后再重新安装。
有些包在conda和pip中的版本是不兼容的，因此应该尽量用conda来安装。如果在conda中没有这个版本，再用pip来安装。

三、conda常用命令¶

常用命令¶

conda create -n learn python=3  // 创建一个名为learn的环境并指定python版本为3(最新版本)
conda activate learn //激活learn环境
conda env list // 列出conda管理的所有环境
conda list // 列出当前环境的所有包
conda search --full-name <package_full_name> //搜索包
conda install requests //安装requests包
conda install -n root conda=4.6 // 将conda降回原来版本
conda remove requests //卸载requets包
conda remove -n learn --all // 删除learn环境及下属所有包
conda update requests 更新requests包
conda env export > environment.yaml  // 导出当前环境的包信息
conda env create -f environment.yaml  // 用配置文件创建新的虚拟环境

1. 验证conda已被安装¶

conda --version

终端上将会以 conda 版本号的形式显示当前安装conda的版本号。如： conda 3.11.0

2. 更新conda至最新版本¶

conda update anaconda #更新anaconda  
conda update conda
conda update --all #更新所有包

执行命令后，conda将会对版本进行比较并列出可以升级的版本。同时，也会告知用户其他相关包也会升级到相应版本。当较新的版本可以用于升级时，终端会显示 Proceed ([y]/n)? ，此时输入 y 即可进行升级。

3. 查看conda帮助信息¶

conda --help
conda -h

4. 卸载conda¶

rm -rf ~/anaconda3

直接删除Anaconda的安装目录即可。

5. 配置¶

anaconda默认会开机启动，为了避免这种情况，我们需要在~/.condarc文件中进行设置。

conda config --set auto_activate_base false

6. 路径及重命名¶

conda env list #可列出conda管理的所有环境

conda不能对环境进行重命名，如果直接将环境所在目录重命名的话：

mv /home/zhangfeng/bin/anaconda3/envs/test /home/zhangfeng/bin/anaconda3/envs/test1 #

使用软件会报错，因为会用到以前的test下的库。只能删除这个环境，重新建一个新环境。

四、管理环境¶

1. 创建新环境¶

conda create --name <env_name> <package_names>

即创建的环境名。建议以英文命名，且不加空格，名称两边不加尖括号“<>”。
即安装在环境中的包名。名称两边不加尖括号“<>”。
如果要安装指定的版本号，则只需要在包名后面以 = 和版本号的形式执行。如： conda create --name python2 python=2.7 ，即创建一个名为“python2”的环境，环境中安装版本为2.7的python。
如果要在新创建的环境中创建多个包，则直接在后以空格隔开，添加多个包名即可。如： conda create -n python3 python=3.5 numpy pandas，即创建一个名为“python3”的环境，环境中安装版本为3.5的python，同时也安装了numpy和pandas。
--name 同样可以替换为 -n 。
默认情况下，新创建的环境将会被保存在 /Users//anaconda3/env 目录下，其中为当前用户的用户名。

2. 切换环境¶

conda activate <env_name>

如果创建环境后安装Python时没有指定Python的版本，那么将会安装与Anaconda版本相同的Python版本，即如果安装Anaconda第2版，则会自动安装Python 2.x；如果安装Anaconda第3版，则会自动安装Python 3.x。
当成功切换环境之后，在该行行首将以“(env_name)”或“[env_name]”开头。其中，“env_name”为切换到的环境名。如：在执行conda activate python2，即切换至名为“python2”的环境，则行首将会以(python2)开头。

3. 退出环境¶

source deactivate

当执行退出当前环境后，原本行首以“(env_name)”或“[env_name]”开头的字符将不再显示。

4. 显示已创建环境¶

conda info --envs
conda info -e
conda env list

结果中星号“*”所在行即为当前所在环境。Linux系统中默认创建的环境名为“base”。

5. 复制环境¶

conda create --name <new_env_name> --clone <copied_env_name>

即为被复制/克隆环境名。环境名两边不加尖括号“<>”。
即为复制之后新环境的名称。环境名两边不加尖括号“<>”。
conda create --name py2 --clone python2，即为克隆名为“python2”的环境，克隆后的新环境名为“py2”。此时，环境中将同时存在“python2”和“py2”环境，且两个环境的配置相同。

6. 删除环境¶

conda remove --name <env_name> --all

为被删除环境的名称。环境名两边不加尖括号“<>”。

五、管理软件包¶

1. 查找可供安装的包版本¶

1.1 精确查找¶

conda search --full-name <package_full_name>

--full-name 为精确查找的参数。
是被查找包的全名。包名两边不加尖括号“<>”。
例如： conda search --full-name python 即查找全名为“python”的包有哪些版本可供安装。

1.2 模糊查找¶

conda search <text>

<text> 是查找含有此字段的包名。此字段两边不加尖括号“<>”。
例如： conda search py 即查找含有“py”字段的包，有哪些版本可供安装。

2. 获取当前环境中已安装的包信息¶

conda list

执行上述命令后将在终端显示当前环境已安装包的包名及其版本号。

3. 安装包¶

3.1 在指定环境中安装包¶

conda install --name <env_name> <package_name>

即将包安装的指定环境名。环境名两边不加尖括号“<>”。
即要安装的包名。包名两边不加尖括号“<>”。
例如： conda install --name python2 pandas 即在名为“python2”的环境中安装pandas包。

3.2 在当前环境中安装包¶

conda install <package_name>

即要安装的包名。包名两边不加尖括号“<>”。
执行命令后在当前环境中安装包。如conda install gdal=2.4.3

3.3 使用pip安装包¶

pip3 install see

当使用 conda install 无法进行安装时，可以使用pip进行安装。例如：see包。

pip 用法：

安裝：pip install PackageName
更新：pip install -U PackageName
移除：pip uninstall PackageName
搜索：pip search PackageName
帮助：pip help
从github安装: pip install git+https://github.com/jkbr/httpie.git

3.4 从http://Anaconda.org安装包¶

当使用 conda install 无法进行安装时，可以考虑从 http://Anaconda.org 中获取安装包的命令，并进行安装。

在浏览器中输入： http://anaconda.org
在SEARCH PACKAGES搜索框中输入要安装的包名，然后点击右边“放大镜”标志进行检索。
搜索结果中有数以千计的包可供选择，此时点击“Downloads”可根据下载量进行排序，最上面的为下载最多的包。
选择满足需求的包或下载量最多的包，点击包名。
“To install this package with conda run one of the following:”下方有安装命令，选择其中一个并粘贴在终端中执行。
完成安装。

4. 卸载软件包¶

4.1 卸载其它环境中的包¶

conda remove --name <env_name> <package_name>

即卸载包所在指定环境的名称。环境名两边不加尖括号“<>”。
即要卸载包的名称。包名两边不加尖括号“<>”。
例如：conda remove --name python2 pandas 即卸载名为“python2”中的pandas包。

4.2 卸载当前环境中的包¶

conda remove <package_name>

即要卸载包的名称。包名两边不加尖括号“<>”。
执行命令后即在当前环境中卸载指定包。
例如： conda remove pandas 即在当前环境中卸载pandas包。

5. 更新包¶

5.1 更新所有包¶

conda update --all
conda upgrade --all

建议：在安装Anaconda之后执行上述命令更新Anaconda中的所有包至最新版本，便于使用。

5.2 更新指定包¶

conda update <package_name>
conda upgrade <package_name>

为指定更新的包名。包名两边不加尖括号“<>”。
更新多个指定包，则包名以空格隔开，向后排列。
如： conda update pandas numpy matplotlib 即更新pandas、numpy、matplotlib包。

六、debug¶

conda install 一直在Solving environment卡死

可能是通道的原因，换一个通道 conda update conda -c conda-forge试试.
删除安装缓存。conda info找到缓存地址，在package cache行。然后进行删除：

rm /home/disk2/xs/anaconda3/pkgs
rm /home/disk2/xs/.conda/pkgs

七、案例¶

创建R环境并安装R软件及R包¶

conda create -n R # 创建一个R的环境
conda activate R #激活R环境
conda install -c conda-forge r-base #不能同时加 r-essentials，会导致环境冲突
conda update -c conda-forge r-base # 如果没有安装最新版本，用此命令尝试更新
conda install -c conda-forge r-packagename #安装r包

注意R包的名称前要加r-，如需要安装ggplot2，命令为conda install -c conda-forge r-ggplot2
有些r包有时就是安装不了。这时候可以直接进入R命令行，直接在R里面装。

创建rnaseq环境¶

conda create -n rnaseq python=3
conda install -c bioconda fastqc multiqc trim-galore 
conda install -c bioconda samtools stringtie STAR

# 验证相关软件是否安装
fastqc --version
multiqc --version
trim_galore --version
samtools --version
stringtie --version
STAR --version

single cell RNA-seq全流程复现HCC文章

2021-09-06T10:20:00+08:00

Single-cell RNA sequencing shows the immunosuppressive landscape and tumor heterogeneity of HBV-associated hepatocellular carcinoma是2021年6月刚发表在NC上的文章，主要是利用单细胞转录组分析HCC患者的肿瘤异质性和免疫相关内容。本文主要复现这篇文章的生物信息分析方面的结果，包括上游的比对和定量，下游的异质性和细胞互作等内容。

The upstream analyses of Single cell RNA-seq data¶

Data collection¶

The sequence data for 8 liver carcinoma was deposited in NCBI with accession code of SRP318499. We download the bam files of 8 samples using ascp software. The size of 8 samples was listed below:

du -sh rawData/*bam
## 56G  rawData/095_possorted_genome_bam.bam
## 23G  rawData/104_possorted_genome_bam.bam
## 22G  rawData/106_possorted_genome_bam.bam
## 51G  rawData/114_possorted_genome_bam.bam
## 45G  rawData/119_possorted_genome_bam.bam
## 22G  rawData/713_possorted_genome_bam.bam
## 27G  rawData/725_possorted_genome_bam.bam
## 53G  rawData/740_possorted_genome_bam.bam

Gene expression quantification¶

We then convert bam files to fastq format, count gene expression for each samples, and aggregate 8 count matrices to one matrix. Take sample with 119 id for example:

# 1. the barcoded BAM files are covered to FASTQ files with the 10x Genomics bamtofastq tool
bamtofastq --nthreads=8 119_possorted_genome_bam.bam 119

# 2. perform read alignment, UMI counting,
cd 119
cellranger count --id=119 --transcriptome=/media/zhangfeng/myData/reference/refdata-gex-GRCh38-2020-A --fastqs=trial5d_count_new_0_1_HG3WLDMXX --chemistry=SC3Pv2 # note: the default parameter of chemistry would throw error, the SC3Pv2 should be defined

# 3. use the cellranger aggr pipeline to aggregate outputs from multiple runs of cellranger count, normalize runs to the same effective sequencing depth
cellranger aggr --id=aggr --csv=libraries.csv

We can review the summary from the web_summary.html. The aggr analysis detected a error : Low Post-Normalization Read Depth of 19.8% , which means that there may be large differences in sequencing depth across the input libraries. So we have to quality control for each samples and then combine them together.

Quality control¶

Firstly, I check the quality of features, counts, and the percentage of mitochondrial genes

library(Seurat)
load("cluster/2_qc/rawSeurats.RData")
dim(rawSeurats);

## [1] 26428 41630

table(rawSeurats@active.ident)

##
##   095   104   106   114   119   713   725   740
##  8388   540   757 11855  7535  1368  2464  8723

VlnPlot(rawSeurats,features = c("nFeature_RNA"))

VlnPlot(rawSeurats,features = c("nCount_RNA"))

VlnPlot(rawSeurats,features = c("percent.mt"))

I perform the qulity control analysis as proposed by Current best practices in single‐cell RNA‐seq analysis: a tutorial, which was published by Luecken at 2019. The distributions of these QC covariates are examined for outlier peaks that are filtered out by thresholding. After quality control, the features, counts, and the percentage of mitochondrial genes are checked again:

load("cluster/2_qc/cleanSeurats.RData")
dim(cleanSeurats);

## [1] 26428 17371

table(cleanSeurats@active.ident)

##
##  095  104  106  114  119  713  725  740
## 3719  406  435 5721 3030  894 1439 1727

VlnPlot(cleanSeurats,features = c("nFeature_RNA"))

VlnPlot(cleanSeurats,features = c("nCount_RNA"))

VlnPlot(cleanSeurats,features = c("percent.mt"))

The plots show that the outlier value are deleted.

Cluster and annotation¶

The Seurat package then is used to preform PCA, reduction, cluster, UMAP and TSNE analysis

1. PCA¶

I conduct the PCA analysis for all cells :

load("cluster/3_cluster/cleanSeurats_standard.RData")
DimPlot(object = cleanSeurats, reduction = "pca",label =T)

DimPlot(object = cleanSeurats, reduction = "pca",label =T,group.by = "orig.ident")

The PCA plots show that maybe batch effect exist. The CCA method is used to adjust batch effect. I then perform the CCA analysis to adjust the batch effect:

load("cluster/3_cluster/cleanSeurats_adjusted.RData")
DimPlot(object = cleanAdjustSeurats, reduction = "tsne",label =T,group.by = "orig.ident")

DimPlot(object = cleanAdjustSeurats, reduction = "umap",label =T,group.by = "orig.ident")

Due to the heterogeneity patients with liver cancer, the tumor cell across cancer samples should be independent each other. However, these cells are mixed. I think that the batch effect is over-estimated. Finally, I decide to run standard workflow as Seurat proposed. The step of batch effect adjustion analysis is skipped.

2. Reduction and TSNE analysis¶

The UMAP and TSNE plot are drew below:

load("cluster/3_cluster/cleanSeurats_standard.RData")
DimPlot(object = cleanSeurats, reduction = "tsne",label =T)

DimPlot(object = cleanSeurats, reduction = "tsne",label =T,group.by = "orig.ident")

The patients with tumor cell are split, which is coincident with our common sense. However, in comparision with TSNE, the UMAP is better, which can clearly show the difference between case identify, and infiltrating immnue cells.

3. UMAP analysis¶

DimPlot(object = cleanSeurats, reduction = "umap",label =T)

DimPlot(object = cleanSeurats, reduction = "umap",label =T,group.by = "orig.ident")

Taken together, we used UMAP to show the cluster of different cell types. The SingleR and markers are combined to define different cell types.

load("cluster/3_cluster/cleanSeurats_anno.RData")
DimPlot(object = cleanSeurats, reduction = "umap",label =T,group.by = "celltype_fine")

The downstream analyses of Single cell RNA-seq data¶

Cell type heterogeneity¶

We find that the cells from the same case are tended to cluster together:

DimPlot(object = cleanSeurats, reduction = "umap",label =T,group.by = "celltype_fine")

However, different HCC case have different proportion of immune cells. The proportion of different cells in 8 HCC cases are calculated :

library(data.table)
library(ggplot2)
library(ggpubr)
library(stringr)
load("cluster/3_cluster/cleanSeurats_anno.RData")

############### T cells and macrophages
# cell proportion in cases
anno = cleanSeurats@meta.data %>% setDT()
cellPro = anno[,.(totalNum=.N,
                  Non_M2=sum(celltype_fine=="Macrophages")/.N,
                  M2=sum(celltype_fine=="M2-macrophages")/.N,
                  CD8=sum(celltype_fine=="CD8 cells")/.N,
                  CD4=sum(celltype_fine=="CD4 cells")/.N,
                  B_cells=sum(celltype_fine=="B cells")/.N,
                  NK=sum(celltype_fine=="NK cells")/.N,
                  Tumor=sum(str_detect(celltype_fine,"^#"))/.N,
                  Other=sum(celltype_fine%in%c("Endothelial cells","Treg cells"))/.N),by="orig.ident"]
rowSums(cellPro[,-c(1:2)]) # equal to 1

## [1] 1 1 1 1 1 1 1 1

y=as.matrix(t(cellPro[,-c(1:2)]))
barplot(y,xlim=c(0, ncol(y) + 4),
        col=c("red","black","blue","yellow","gray","green","white","purple"),
        names.arg=cellPro$orig.ident,
        legend.text=TRUE,
        args.legend=list(
          x=ncol(y) + 4,
          y=max(colSums(y)),
          bty ="n"
        ))

T cells and macrophages¶

The correlation between proportion of T cells and macrophages in cases are calculated:

# correlation
ggplot(cellPro, aes(x = M2, y = CD8)) +
  geom_point() +
  stat_smooth(method = "lm", col = "blue")+
  stat_cor(label.x = 0, label.y = -0.2) +
  labs(title="scRNA-seq cohort", x="M2 macrophages proportion", y="CD8T proportion")+
  theme(plot.title = element_text(hjust = 0.5))

To compared with bulk RNA-seq , the immune cell deconvolution analysis is performed, and the correaltion relationship is represent by linear model :

# tcga immune cell
lihc = read.csv("cluster/6_prognosis/TCGA_LIHC_TME_results.csv",head=T)
ggplot(lihc, aes(x = Macrophages.M2, y = T.cells.CD8)) +
  geom_point() +
  stat_smooth(method = "lm", col = "blue")+
  stat_cor(label.x = 0.4, label.y = 0.4)+
  labs(title="TCGA LIHC cohort", x="M2 macrophages proportion", y="CD8T proportion")+
  theme(plot.title = element_text(hjust = 0.5))

We also detect inverse correlation between the proportions of infiltrating T cells and macrophages.

Immunosuppressive marker expression of TAMs¶

We examined the expression of reported immunosuppressive genes in TAMs:

load("cluster/3_cluster/cleanSeurats_anno.RData")
FeaturePlot(cleanSeurats, features = c("LAIR1","HAVCR2","LGALS9","VSIR"))

DimPlot(object = cleanSeurats, label =T,group.by = "celltype_fine")

These markers are enriched in TAM cells. In addition, the pattern of CD163 (M2 macrophages marker) expression and LAIR1 is similar:

cellKeep = colnames(cleanSeurats)[cleanSeurats@meta.data$celltype_main=="Macrophages"]
VlnPlot(subset(cleanSeurats,cells=cellKeep), features = c("CD163", "LAIR1", "CD68","HAVCR2","LGALS9","VSIR"))

These suggest the immunosuppressive function of TAMs might be exerted via LAIR1. In addition, the high expression of LAIR1 and HAVCR2 is significantly associated with poorer overall or disease-free survival of HCC patients.

The overall survival for LAIR1 , and for HAVCR2 .

The disease-free survival for LAIR1 , and for HAVCR2 .

Cell trajectory¶

library(monocle3)
library(SeuratWrappers)
load("cluster/3_cluster/cleanSeurats_anno.RData")
get_earliest_principal_node <- function(cds, cluster="1"){
  cell_ids <- which(colData(cds)[, "seurat_clusters"] == cluster)

closest_vertex <- cds@principal_graph_aux[["UMAP"]]$pr_graph_cell_proj_closest_vertex
  closest_vertex <- as.matrix(closest_vertex[colnames(cds), ])
  root_pr_nodes <-  igraph::V(principal_graph(cds)[["UMAP"]])$name[as.numeric(names(which.max(table(closest_vertex[cell_ids,]))))]

root_pr_nodes
}

cds <- as.cell_data_set(cleanSeurats)
cds <- estimate_size_factors(cds)
cds@rowRanges@elementMetadata@listData[["gene_short_name"]] <- rownames(cleanSeurats[["RNA"]])

We do the single cell trajectory for CD8 population, and observed that gradual transition of CD8 T cells towards the subpopulations with exhaustion status was indicated by the upregulation of PDCD1 (PD-1) and TIGIT (T cell immunoreceptor with Ig and ITIM domains).

cds_subset <- cds[, colData(cds)$seurat_clusters%in% c(10,16)]
cds_subset <- cluster_cells(cds_subset)
cds_subset <- learn_graph(cds_subset)
cds_subset <- order_cells(cds_subset,root_pr_nodes=get_earliest_principal_node(cds_subset,cluster=10))
# for CD8
plot_cells(cds_subset,
           label_cell_groups=T,
           color_cells_by = "pseudotime",
           show_trajectory_graph=F)

plot_cells(cds_subset,
           label_cell_groups=T,
           genes=c("PDCD1","TIGIT"),
           show_trajectory_graph=F)

We find that exhaustion status was indicated by the downregulation of FCGR3A (activation marker) and upregulation of KLRC1 (exhaustion marker).

# for NK cells
cds_subset <- cds[, colData(cds)$seurat_clusters%in% c(15)]
cds_subset <- cluster_cells(cds_subset)
cds_subset <- learn_graph(cds_subset)
cds_subset <- order_cells(cds_subset,root_pr_nodes=get_earliest_principal_node(cds_subset,cluster=15))

plot_cells(cds_subset,
           label_cell_groups=T,
           color_cells_by = "pseudotime",
           show_trajectory_graph=F)

plot_cells(cds_subset,
           label_cell_groups=T,
           genes=c("FCGR3A","KLRC1"),
           show_trajectory_graph=F)

We find the transition toward a more immunosuppressive state for Treg cells, as indicated by the enriched PDCD1 expression.

cds_subset <- cds[, colData(cds)$seurat_clusters%in% c(5,29)]
cds_subset <- cluster_cells(cds_subset)
cds_subset <- learn_graph(cds_subset)
cds_subset <- order_cells(cds_subset,root_pr_nodes=get_earliest_principal_node(cds_subset,cluster=5))
plot_cells(cds_subset,
           label_cell_groups=T,
           color_cells_by = "pseudotime",
           show_trajectory_graph=F)

plot_cells(cds_subset,
           label_cell_groups=T,
           genes=c("PDCD1"),
           show_trajectory_graph=F)

For macrophages, a dynamic transition towards more immunosuppressive M2 macrophages featured by CD163 and LAIR1 expression.

# for TAMs
cds_subset <- cds[, colData(cds)$seurat_clusters%in% c(3,9,14,18,21)]
cds_subset <- cluster_cells(cds_subset)
cds_subset <- learn_graph(cds_subset)
cds_subset <- order_cells(cds_subset,root_pr_nodes=get_earliest_principal_node(cds_subset,cluster=9))

plot_cells(cds_subset,
           label_cell_groups=T,
           color_cells_by = "pseudotime",
           show_trajectory_graph=F)

plot_cells(cds_subset,
           label_cell_groups=T,
           genes=c("CD163","LAIR1"),
           show_trajectory_graph=F)

Regarding the CD4 T cells, they were mainly type 1 (Th1) and type 2T helper (Th2) cells, as indicated by their respective enrichment in the expression of STAT4 (cell cluster 11) and GATA3 (cell cluster 12).

cds_subset <- cds[, colData(cds)$seurat_clusters%in% c(1,2,6,17)]
cds_subset <- cluster_cells(cds_subset)
cds_subset <- learn_graph(cds_subset)
cds_subset <- order_cells(cds_subset,root_pr_nodes=get_earliest_principal_node(cds_subset,cluster=2))
plot_cells(cds_subset,
           label_cell_groups=T,
           color_cells_by = "pseudotime",
           show_trajectory_graph=F)

plot_cells(cds_subset,
           label_cell_groups=T,
           genes=c("STAT4","GATA3"),
           show_trajectory_graph=F)

Immune checkpoints¶

Immune checkpoints function as “brakes” on T cell immune responses resulting in the weakening of T cell attack and immune tolerance or escape. We examined the cell-cell interaction status in different cell, and the co-stimulatory and co-inhibitory checkpoints in shaping the immunosuppressive andscape in HCC are estimated :

mypvals <- read.delim("cluster/7_interaction/out/pvalues.txt", check.names = FALSE)
mymeans <- read.delim("cluster/7_interaction/out/means.txt", check.names = FALSE)
costimulatory <- grep("ICOS|ICOSLG|TNFRSF4|TNFRSF8|TNFRSF9|TNFRSF18|CD40LG|CD40|CD27|CD70|CD28|CD80|CD86|CD226|PVR|NECTIN1|NECTIN2|NECTIN3|NECTIN4|TNFSF14|TNFRSF14",
                      mymeans$interacting_pair,value = T)
coinhibitory <- grep("PDCD1|CD274|PDCD1LG2|HAVCR2|LGALS9|LAIR1|PTPN6|PTPN11|CTLA4|CD80|CD86|TIGIT|PVR|NECTIN1|NECTIN2|NECTIN3|NECTIN4|CD160|BTLA|TNFRSF14",
                     mymeans$interacting_pair,value = T)

# plot for co-stimulatory
meansdf = mymeans %>% dplyr::filter(interacting_pair %in% costimulatory) %>%
  dplyr::select(c("interacting_pair","CD4 cells|Tumor","CD8 cells|Tumor","Treg cells|Tumor","NK cells|Tumor",
                  "CD4 cells|Macrophages","CD8 cells|Macrophages","Treg cells|Macrophages","NK cells|Macrophages"))  %>%
reshape2::melt()
colnames(meansdf)<- c("interacting_pair","CC","means")
meansdf$joinlab <- paste0(meansdf$interacting_pair,"_",meansdf$CC)

pvalsdf = mypvals %>% dplyr::filter(interacting_pair %in% costimulatory)%>%
  dplyr::select(c("interacting_pair","CD4 cells|Tumor","CD8 cells|Tumor","Treg cells|Tumor","NK cells|Tumor",
                  "CD4 cells|Macrophages","CD8 cells|Macrophages","Treg cells|Macrophages","NK cells|Macrophages"))%>%
reshape2::melt()
colnames(pvalsdf)<- c("interacting_pair","CC","pvals")
pvalsdf$joinlab <- paste0(pvalsdf$interacting_pair,"_",pvalsdf$CC)

ccInter <- merge(pvalsdf,meansdf,by = "joinlab")
summary(ccInter$means)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max.
##  0.0000  0.0160  0.0765  0.1538  0.2085  1.1570

ccInter %>% dplyr::filter(means > 0.1) %>%
  ggplot(aes(CC.x,interacting_pair.x) )+
  geom_point(aes(size=means,colour= -log10(pvals+0.0001))) +
  scale_color_continuous(name=c("-log10(P-value)"),low = "black", high = "red")+
  theme_bw()+ labs(x="",y="")+
  theme(axis.text.x = element_text(angle = -45,hjust = -0.1,vjust = 0.8))

# plot for co-inhibitory
meansdf = mymeans %>% dplyr::filter(interacting_pair %in% coinhibitory) %>%
  dplyr::select(c("interacting_pair","CD4 cells|Tumor","CD8 cells|Tumor","Treg cells|Tumor","NK cells|Tumor",
                  "CD4 cells|Macrophages","CD8 cells|Macrophages","Treg cells|Macrophages","NK cells|Macrophages"))  %>%
reshape2::melt()
colnames(meansdf)<- c("interacting_pair","CC","means")
meansdf$joinlab <- paste0(meansdf$interacting_pair,"_",meansdf$CC)

pvalsdf = mypvals %>% dplyr::filter(interacting_pair %in% coinhibitory)%>%
  dplyr::select(c("interacting_pair","CD4 cells|Tumor","CD8 cells|Tumor","Treg cells|Tumor","NK cells|Tumor",
                  "CD4 cells|Macrophages","CD8 cells|Macrophages","Treg cells|Macrophages","NK cells|Macrophages"))%>%
reshape2::melt()
colnames(pvalsdf)<- c("interacting_pair","CC","pvals")
pvalsdf$joinlab <- paste0(pvalsdf$interacting_pair,"_",pvalsdf$CC)

ccInter <- merge(pvalsdf,meansdf,by = "joinlab")
summary(ccInter$means)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max.
## 0.00000 0.02775 0.10650 0.17020 0.23875 1.15700

ccInter%>% dplyr::filter(means > 0.1) %>%
  ggplot(aes(CC.x,interacting_pair.x) )+
  geom_point(aes(size=means,colour= -log10(pvals+0.0001))) +
  scale_color_continuous(name=c("-log10(P-value)"),low = "black", high = "red")+
  theme_bw()+ labs(x="",y="")+
  theme(axis.text.x = element_text(angle = -45,hjust = -0.1,vjust = 0.8))

We identified a prominent co-inhibitory signal via the TIGIT-NECTIN2 axis in T cells and antigen-presenting cells (APCs, i.e. macrophages and tumor cells):

FeaturePlot(cleanSeurats, features = c("TIGIT","NECTIN2"))

Cell interactions¶

We evaluated the degree of cell-cell communication according to different ligand-receptor relationships. The tumor cells contributing ligands, tumor cells contributing receptors, TAMs contributing ligands and TAMs contributing receptors are presented as :

# plot for tumor ligand
meansdf = mymeans %>% dplyr::select("interacting_pair",starts_with("Tumor"))
meansdf = meansdf[!duplicated(meansdf$interacting_pair),]
row.names(meansdf) = meansdf$interacting_pair
meansdf = as.matrix(meansdf[,-1])
pvalsdf = mypvals[!duplicated(mypvals$interacting_pair),]
row.names(pvalsdf) = pvalsdf$interacting_pair
pvalsdf = pvalsdf[row.names(meansdf),colnames(meansdf)]
meansdf = meansdf[apply(pvalsdf,1,function(x) { sum(x<0.05)>3}),]
pheatmap::pheatmap(meansdf)

# plot for tumor receptor
meansdf = mymeans %>% dplyr::select("interacting_pair",ends_with("Tumor"))
#meansdf = meansdf[!str_detect(meansdf$interacting_pair,pattern=" "),]
meansdf = meansdf[!duplicated(meansdf$interacting_pair),]
row.names(meansdf) = meansdf$interacting_pair
meansdf = as.matrix(meansdf[,-1])
pvalsdf = mypvals[!duplicated(mypvals$interacting_pair),]
row.names(pvalsdf) = pvalsdf$interacting_pair
pvalsdf = pvalsdf[row.names(meansdf),colnames(meansdf)]
meansdf = meansdf[apply(pvalsdf,1,function(x) { sum(x<0.05)>3}),]
pheatmap::pheatmap(meansdf)

#plot for TAM ligand
meansdf = mymeans %>% dplyr::select("interacting_pair",starts_with("Macrophages"))
#meansdf = meansdf[!str_detect(meansdf$interacting_pair,pattern=" "),]
meansdf = meansdf[!duplicated(meansdf$interacting_pair),]
row.names(meansdf) = meansdf$interacting_pair
meansdf = as.matrix(meansdf[,-1])
pvalsdf = mypvals[!duplicated(mypvals$interacting_pair),]
row.names(pvalsdf) = pvalsdf$interacting_pair
pvalsdf = pvalsdf[row.names(meansdf),colnames(meansdf)]
meansdf = meansdf[apply(pvalsdf,1,function(x) { sum(x<0.05)>3}),]
pheatmap::pheatmap(meansdf)

# plot for TAM receptor
meansdf = mymeans %>% dplyr::select("interacting_pair",ends_with("Macrophages"))
#meansdf = meansdf[!str_detect(meansdf$interacting_pair,pattern=" "),]
meansdf = meansdf[!duplicated(meansdf$interacting_pair),]
row.names(meansdf) = meansdf$interacting_pair
meansdf = as.matrix(meansdf[,-1])
pvalsdf = mypvals[!duplicated(mypvals$interacting_pair),]
row.names(pvalsdf) = pvalsdf$interacting_pair
pvalsdf = pvalsdf[row.names(meansdf),colnames(meansdf)]
meansdf = meansdf[apply(pvalsdf,1,function(x) { sum(x<0.05)>3}),]
pheatmap::pheatmap(meansdf)

In general, HCC tumor cells frequently interacted with immune cells.

Subclonal heterogeneity¶

The unsupervised hierarchical clustering of LCSC markers is shown as _cluster" />. All cells are split into 7 clusters, one of which including 6 cells can be ignored: _umap" /> We find that the modest correlation between case identity and LCSC marker group status.

To further explore the heterogeneity landscape of HCC tumor cell, the CNV status is inferred using infercnv package. The results showed that there are 9 major group of HCC tumor cells. These cells from 9 major group are clustered together according to their case identify.

Idea¶

to research molecular mechanism of key gene based on conditional knockout mice, regrading their basic structure and function, novelty, and unique expressed in certain cell.
using new technology to sequence the HCC sample, such as spatial single RNA-seq technology.

miRNA的靶基因预测详细教程

2020-11-11T10:20:00+08:00

最近发现很多人对预测miRNA的靶基因感兴趣，因此分享一下这方面的知识。通过生信算法预测miRNA靶基因的网站有很多，下面例举3个常见网站，并给出详细使用教程。

一般来说，预测网站的假阳性会比较高，我们可以同时用这些网站进行预测，然后取交集来提高准确率。下面以miR-320为例进行说明。

miRTarBase ¶

以hsa-miR-320a为例，查询后的结果如下图所示：

其中值得注意的是：所有的相互作用对来源于6种证据，有reporter assay, western blot, qPCR, Microarray, NGS, PSILAC。前面三种证据的可靠较强。

miRDB ¶

miRDB的可靠性则通过Target Score来体现，从高到低进行排序。点击相对应的Details可以查看相匹配的位点。

TargetScan ¶

TargetScan 则通过搜索mRNA的保守 8mer, 7mer, 和6mer 位点来与miRNA相匹配。然后按一个总的得分进行排序。

批量分析¶

除了单个miRNA输入查询外，这3个网站同时也提供了数据下载服务。我们可以把miRNA-mRNA相互作用的关系对下载下来，然后用R语言处理一下，就可以批量分析miRNA的靶基因了。

下面是针对人类的预测分析：

rm(list=ls());options(stringsAsFactors=F)
library(stringr)
library(biomaRt)

miRNAs="miR-320a";species="hsa"
source=paste0(species,'-',miRNAs)
miRTarBase = read.csv("miRTarBase_MTI.csv",head=T)
miRDB = read.table("miRDB_v5.0_prediction_result.txt",head=F)
targetscan = read.table("Predicted_Targets_Context_Scores.default_predictions.txt",head=T,sep="\t")


## miRTarBase
miRTarBase_mRNA = miRTarBase[str_detect(miRTarBase$miRNA,paste0('^',source,'$|^',source,'-[35]p')),]

## miRDB
miRDB_mRNA = miRDB[str_detect(miRDB$V1,paste0('^',source,'$|^',source,'-[35]p')),] 
ensembl = useEnsembl(biomart="ensembl", dataset="hsapiens_gene_ensembl")
miRDBgene = getBM(attributes = c( 'refseq_mrna',"hgnc_symbol"), filters = 'refseq_mrna', values = unique(miRDB_mRNA$V2), mart = ensembl)
miRDBgene = miRDBgene[miRDBgene$hgnc_symbol!="",]
miRDB_mRNA = merge(miRDB_mRNA,miRDBgene,by.x="V2",by.y="refseq_mrna",sort=F)

## TargetScan
targetscan_mRNA = targetscan[str_detect(targetscan$miRNA,paste0('^',source,'$|^',source,'-[35]p')),]

## merge
res = merge(miRTarBase_mRNA,targetscan_mRNA,by.x="Target.Gene",by.y="Gene.Symbol")
res = merge(res,miRDB_mRNA,by.x="Target.Gene",by.y="hgnc_symbol")
write.csv(res,file="miRNAtargetGene.csv",row.names = F,col.names = F)

这个代码还不完善，之后有机会把它们做成一个R软件包，更方便使用。

生物信息最佳入门实践路线图

2020-10-10T10:20:00+08:00

生物信息最佳入门实践路线图¶

介绍¶

生信是一门交叉学科，要学习的东西非常多。本人从2011年开始学习生信，深知它的博大精深。因此，给大家推荐一些学习资料，可以快速入门。由于我是一位喜欢阅读多过视听学习的人，所以书籍都是亲自阅读过的，但是视频并没有认真看过。

下面这张图展示了生信学习的路线图。一开始可能就是学习一门编程语言，首选R语言，然后是python。其次如果你要处理大量测序数据，你得学习linux操作系统。最后最重要的一步是：利用项目不断进行实践。当然最终目标还是为了理解生物学，利用你的生物学知识来解释分析结果对生物学的意义。

其实生信这一领域或者说计算机这一领域最重要的一个能力就是学习。学习怎么搜索问题和答案并亲自去尝试。得学会怎么高效搜索，可以google, bing, QQ群，论坛（如stackoverflow），老师和同学等。大部分问题都可以在网上得到解决，相信我，你肯定不是第一个碰到问题的人。然后是在QQ或论坛去请教，当然如果你身边有生信的高手请教可以节省大量时间。

所有资料我都尽量用开源免费的，这样省掉很多学习成本。另外有一些收集的pdf书籍放在百度云上，链接，提取码：bioT。如果失效了请在最下方的评论区提醒一下。

R语言¶

如果你想快速入门深信，请必学R语言。这是一门开源的编程语言，主要优势在于统计和绘图，以及极其丰富的软件包。

R语言资料¶

如果你是一位喜欢视听学习这种方式的，请参考生信技能树的生信人应该这样学R语言。
如果你是一位喜欢阅读学习这种方式，请参考R for beginners和R现代统计图形。如果你有余力还可以学习R语言实战及ggplot2:数据分析与图形艺术等深入内容。
最后进行这20个R语言习题进行实践，加深印象。如果你没有时间，而且入门资料都能看得懂，也可以跳过这步，在项目实践阶段结合自己课题进行练习。

linux¶

如果要对测序后的原始数据进行上游分析的话，你一定得学会Linux。如果你不用处理多组学数据的原始序列，那么这个可以跳过。

入门视频：https://www.bilibili.com/video/av28813815
书籍：鸟哥的linux私房菜是繁体字，也有转成简体中文版本。
linux 实践：http://www.bio-info-trainee.com/2900.html

python¶

python是一门胶水语言，基本上什么都干，尤其是在机器学习方面。目前很多机器学习和神经网络的开源项目都是用python写的，如果你有兴趣往这方面发展。python是非学不可的。

编辑入门：笨办法学 Python
机器学习入门：python机器学习基础教程（Andreas C.Muller and Sarah Guido著，张亮译）
神经网络入门：[Python深度学习（Francois Chollet 著，张亮译）

项目实践¶

目前没找到好的开源项目进行实践，这个跟课题组的研究内容相关。不过下面列出一些组学的内容可以用来实践。

转录组：https://www.bilibili.com/video/av28453557
单细胞组：https://www.bilibili.com/video/av38741055
肿瘤免疫的生信分析，分享一些文献进行具体实战。链接：https://pan.baidu.com/s/10nUWgVR4SRT83Qwmwuy0pg 提取码：qapd

使用不同方法规一化、PCA和批次效应矫正的差异

2018-12-03T10:20:00+08:00

1）规一化表达矩阵¶

转录组差异表达分析时，需要对表达矩阵进行规一化。比较常用的方法有log2和DESeq2包的vst和rlog方法。那么它们有什么不一样呢？我们拿DESeq2的示例数据进行比较。

library(RNAseqFlow)
library(DESeq2)
library(corrplot)
input <- createCountPhe() # 得到表达矩阵和表型信息
expr <- input[[1]]
phe <- input[[2]]
dds <- create_DEseq(count_data=expr,col_data=phe,design_names = "condition+type",group_name ="condition",ref_level = "untreated") # 创建DESeqDataSet对象

## [1] "The id order between gene count file and phenotype file is identical without modification!"

dds

## class: DESeqDataSet
## dim: 10089 7
## metadata(1): version
## assays(4): counts mu H cooks
## rownames(10089): FBgn0000008 FBgn0000017 ... FBgn0261574 FBgn0261575
## rowData names(26): baseMean baseVar ... deviance maxCooks
## colnames(7): treated1 treated2 ... untreated3 untreated4
## colData names(3): condition type sizeFactor

norm <- log2(expr+1) #直接用log2进行规一化
norm[1:5,1:5]

##              treated1  treated2  treated3 untreated1 untreated2
## FBgn0000008  7.139551  6.475733  6.149747   6.539159   7.339850
## FBgn0000017 12.599448 11.585432 11.703471  12.187661  13.089285
## FBgn0000018  9.497852  8.228819  8.271463   9.189825   9.573647
## FBgn0000024  3.459432  3.000000  2.584963   3.459432   3.584963
## FBgn0000032 10.730470  9.445015  9.566054  10.498849  10.743151

ntd <- normTransform(dds) # DESeq中的函数
assay(ntd)[1:5,1:5]

##              treated1  treated2  treated3 untreated1 untreated2
## FBgn0000008  6.436242  6.865379  6.410556   6.354468   6.504519
## FBgn0000017 11.889798 11.978840 11.967612  12.000870  12.247040
## FBgn0000018  8.789322  8.621190  8.534895   9.003332   8.732772
## FBgn0000024  2.830673  3.349813  2.808365   3.290618   2.834907
## FBgn0000032 10.021212  9.838041  9.829949  10.312153   9.901443

normalized_counts = counts(dds, normalized=TRUE)
ntd1 = log2(normalized_counts+1)
ntd1[1:5,1:5]

##              treated1  treated2  treated3 untreated1 untreated2
## FBgn0000008  6.436242  6.865379  6.410556   6.354468   6.504519
## FBgn0000017 11.889798 11.978840 11.967612  12.000870  12.247040
## FBgn0000018  8.789322  8.621190  8.534895   9.003332   8.732772
## FBgn0000024  2.830673  3.349813  2.808365   3.290618   2.834907
## FBgn0000032 10.021212  9.838041  9.829949  10.312153   9.901443

rld <- rlog(dds, blind=FALSE) # DESeq中的函数
assay(rld)[1:5,1:5]

##              treated1  treated2  treated3 untreated1 untreated2
## FBgn0000008  6.499251  6.678569  6.494630   6.469361   6.527893
## FBgn0000017 11.933399 12.001811 11.993193  12.018453  12.208070
## FBgn0000018  8.758693  8.649608  8.594954   8.900817   8.721092
## FBgn0000024  2.679631  2.713855  2.679046   2.712598   2.679846
## FBgn0000032  9.998801  9.867264  9.861304  10.211745   9.911647

vsd <- vst(dds, blind=FALSE) # DESeq中的函数
assay(vsd)[1:5,1:5]

##              treated1  treated2  treated3 untreated1 untreated2
## FBgn0000008  8.005828  8.192922  7.995328   7.972665   8.034115
## FBgn0000017 11.968753 12.053246 12.042577  12.074190  12.309219
## FBgn0000018  9.322821  9.204951  9.145817   9.477809   9.282788
## FBgn0000024  7.105652  7.182429  7.102606   7.173068   7.106233
## FBgn0000032 10.283202 10.130746 10.124080  10.531172  10.183179

sample <- cbind(log2=norm[,1],ntd=assay(ntd)[,1],rld=assay(rld)[,1],vsd=assay(vsd)[,1])
corr <- cor(sample,use="pairwise.complete.obs")
corrplot(corr, order = "original", type = "upper", mar=c(1,5,5,1),tl.pos="n")
corrplot(corr, add = TRUE, type = "lower", method = "number", order = "original",
         tl.pos="full",cl.pos="n")

从数值来看，不同方法相差不大。从相关性来看，不同方法规一化的数据差异也不大，以第一个样本为例，相关性都达到0.94以上。因此，不同规一化方法差别不是特别大，选择其中一个就好。

不同方法的运行速度：

log2 方法最简单，最方便。
normTransform 与log2类似，但是它会先进行size factor的规一化再log2。需要首先转成DESeqDataSet类。
rlog 速度比较慢。需要首先转成DESeqDataSet类。
vst是rlog的快速版。需要首先转成DESeqDataSet类。

2）PCA 展示¶

如果输入数据是DESeqDataSet类，可以使用plotPCA进行展示：

plotPCA(ntd,intgroup=c("condition"))

plotPCA(rld,intgroup=c("condition"))

plotPCA(vsd,intgroup=c("condition"))

如果是log2的结果，可以用plot_PCA展示：

library(bioTools)
library(ggplot2)
plot_PCA(t(norm),phe=input[[2]],group = "condition")

plot_PCA是bioTools包的函数，可用devtools::install_github("Feng-Zhang/bioTools")进行安装。 plot_PCA会使用prcomp函数计算距离矩阵，再用ggplot2画图。prcomp函数的输入文件是data.frame（n*m），行(n)为样本，列(m)为多维变量。prcomp会将多维变量降维到n维。注意如果mat是没有规一化的数据，需要设置其中的参数scale=T，将数据进行规一化。

可以看到rlog和vst的结果是类似的，且比normTransform方法好，因为PC1和PC2加起来解释的方差更大。而log2方法的效果也很不错。因此：

如果使用DESeq2进行相关分析时，推荐使用vst方法。
如果输入文件是一般的表达矩阵，用log2的方法也是不错的选择，不用费劲地转成DESeqDataSet类。
用plot_PCA可以方便地画PCA图。它是用ggplot2画的，返回的结果还可以用ggplot2进一步优化。

3）查看批次效应¶

我们一般用PCA方法来看样本间是否有批次效应。本例中condition是我们关注的因素，但是type不是，它是指不同测序平台类型。我们不希望不同平台的数据对我们所关注的因素造成影响，即不希望不同平台的测序结果有系统性差异。因此可以先用PCA方法查看type对样本的影响。

plotPCA(ntd,intgroup=c("type"))

plotPCA(rld,intgroup=c("type"))

plotPCA(vsd,intgroup=c("type"))

plot_PCA(t(norm),phe=input[[2]],group="type")

我们看到所有方法的规一化数据都表明，不同type之间的样本有明显的差别，即存在显著的批次效应。

4）矫正批次效应¶

1. 用limma去除批次效应¶

model <-  model.matrix(~condition, colData(dds))
mat <- limma::removeBatchEffect(assay(ntd), ntd$type,design = model)
assay(ntd) <- mat
plotPCA(ntd,intgroup=c("condition"))

plotPCA(ntd,intgroup=c( "type"))

mat <- limma::removeBatchEffect(assay(rld), rld$type,design = model)
assay(rld) <- mat
plotPCA(rld, intgroup=c("condition"))

plotPCA(rld, intgroup=c( "type"))

mat <- limma::removeBatchEffect(assay(vsd), vsd$type,design = model)
assay(vsd) <- mat
plotPCA(vsd, intgroup=c("condition"))

plotPCA(vsd, intgroup=c("type"))

可以看到样本可以被condition分开，但不能被type分开，表明矫正成功。

2. 用ComBat去除批次效应¶

library(sva)
model <- model.matrix(~condition,data = phe)
clean_norm <- norm[rowSums(norm)>1 & rowMedians(norm)>1,]
adjusted_expr <- ComBat(dat = clean_norm, batch = as.character(phe$type))

## Found 33 genes with uniform expression within a single batch (all zeros); these will not be adjusted for batch.

plot_PCA(t(adjusted_expr),phe=phe,group = "condition")

plot_PCA(t(adjusted_expr),phe=phe,group = "type")

注意ComBat的输入文件是过滤且规一化的数据，不能是原始count。