参考：生信会客厅
harmony原理
Harmony需要输入低维空间的坐标值（embedding），一般使用PCA的降维结果。Harmony导入PCA的降维数据后，会采用soft k-means clustering算法将细胞聚类。常用的聚类算法仅考虑细胞在低维空间的距离，但是soft clustering算法会考虑我们提供的校正因素。这就好比我们的高考加分制度，小明高考成绩本来达不到A大学的录取分数线，但是他有一项省级竞赛一等奖加10分就够线了。同样的道理，细胞c2距离cluster1有点远，本来不能算作cluster1的一份子；但是c2和cluster1的细胞来自不同的数据集，因为我们期望不同的数据集融合，所以破例让它加入cluster1了。聚类之后先计算每个cluster内各个数据集的细胞的中心点，然后根据这些中心点计算各个cluster的中心点。最后通过算法让cluster内的细胞向中心聚集，实在收敛不了的离群细胞就过滤掉。调整之后的数据重复：聚类—计算cluster中心点—收敛细胞—聚类的过程，不断迭代直至聚类效果趋于稳定。
- 从第三方的评测结果来看，harmony的整合效果可以与seurat的锚点整合方法媲美，并且运行时间方面具有明显优势，不过内存消耗好像没有那么夸张的优秀。(哈哈我的电脑用seurat锚点整合被卡住了，试了一下harmony，时间挺快的，而且电脑也带得动（记录一下学习笔记）
library(harmony)
install_github("immunogenomics/harmony")
library(devtools)
##==准备seurat对象列表==##
library(Seurat)
library(tidyverse)
library(patchwork)
rm(list=ls())
dir = c('Data/GSE139324/GSE139324_SUB/GSM4138110',
'Data/GSE139324/GSE139324_SUB/GSM4138111',
'Data/GSE139324/GSE139324_SUB/GSM4138128',
'Data/GSE139324/GSE139324_SUB/GSM4138129',
'Data/GSE139324/GSE139324_SUB/GSM4138148',
'Data/GSE139324/GSE139324_SUB/GSM4138149',
'Data/GSE139324/GSE139324_SUB/GSM4138162',
'Data/GSE139324/GSE139324_SUB/GSM4138163',
'Data/GSE139324/GSE139324_SUB/GSM4138168',
'Data/GSE139324/GSE139324_SUB/GSM4138169') #GSE139324是存放数据的目录
sample_name <- c('HNC01PBMC', 'HNC01TIL', 'HNC10PBMC', 'HNC10TIL', 'HNC20PBMC',
'HNC20TIL', 'PBMC1', 'PBMC2', 'Tonsil1', 'Tonsil2')
scRNAlist <- list()
for(i in 1:length(dir)){
counts <- Read10X(data.dir = dir[i])
scRNAlist[[i]] <- CreateSeuratObject(counts, project=sample_name[i], min.cells=3, min.features = 200)
scRNAlist[[i]][["percent.mt"]] <- PercentageFeatureSet(scRNAlist[[i]], pattern = "^MT-")
scRNAlist[[i]] <- subset(scRNAlist[[i]], subset = percent.mt < 10)
}
saveRDS(scRNAlist, "scRNAlist.rds")
##==harmony整合多样本==##
library(harmony)
scRNAlist <- readRDS("scRNAlist.rds")
##PCA降维
scRNA_harmony <- merge(scRNAlist[[1]], y=c(scRNAlist[[2]], scRNAlist[[3]], scRNAlist[[4]], scRNAlist[[5]],
scRNAlist[[6]], scRNAlist[[7]], scRNAlist[[8]], scRNAlist[[9]], scRNAlist[[10]]))
scRNA_harmony <- NormalizeData(scRNA_harmony) %>% FindVariableFeatures() %>% ScaleData() %>% RunPCA(verbose=FALSE)
##整合
system.time({scRNA_harmony <- RunHarmony(scRNA_harmony, group.by.vars = "orig.ident")})
# 用户 系统 流逝
# 34.308 0.024 34.324
#user system elapsed
#62.90 4.06 72.08
#降维聚类
scRNA_harmony <- RunUMAP(scRNA_harmony, reduction = "harmony", dims = 1:30)
scRNA_harmony <- FindNeighbors(scRNA_harmony, reduction = "harmony", dims = 1:30) %>% FindClusters()
##作图
#group_by_cluster
plot1 = DimPlot(scRNA_harmony, reduction = "umap", label=T)
#group_by_sample
plot2 = DimPlot(scRNA_harmony, reduction = "umap", group.by='orig.ident')
#combinate
plotc <- plot1+plot2
plotc
练习另外一个数据
#harmonny
library(harmony)
library(Seurat)
library(tidyverse)
library(dplyr)
library(patchwork)
setwd("D:\\genetic_r\\study\\geo\\lesson7")
library(harmony)
scRNA.2=Read10X("D:\\genetic_r\\study\\geo\\lesson7\\GSE152048_BC2.matrix\\BC2")
scRNA.3=Read10X("D:\\genetic_r\\study\\geo\\lesson7\\GSE152048_BC3.matrix\\BC3")
scRNA.2 = CreateSeuratObject(scRNA.2 ,project="sample_2",min.cells = 3, min.features = 200)
scRNA.3 = CreateSeuratObject(scRNA.3 ,project="sample_3",min.cells = 3, min.features = 200)
scRNA_harmony <- merge(scRNA.2, y=c(scRNA.3 ))
Warning message:
In CheckDuplicateCellNames(object.list = objects) :
Some cell names are duplicated across objects provided. Renaming to enforce unique cell names.
scRNA_harmony <- NormalizeData(scRNA_harmony) %>% FindVariableFeatures() %>% ScaleData() %>% RunPCA(verbose=FALSE)
Performing log-normalization
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Calculating gene variances
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Calculating feature variances of standardized and clipped values
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Centering and scaling data matrix
|============================================================| 100%
#开始整合
system.time({scRNA_harmony <- RunHarmony(scRNA_harmony, group.by.vars = "orig.ident")})
Harmony 1/10
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Harmony 2/10
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Harmony 3/10
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Harmony 4/10
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Harmony 5/10
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Harmony 6/10
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Harmony converged after 6 iterations
Warning: Invalid name supplied, making object name syntactically valid. New object name is Seurat..ProjectDim.RNA.harmony; see ?make.names for more details on syntax validity
用户 系统 流逝
67.39 5.28 88.92
降维聚类
scRNA_harmony <- FindNeighbors(scRNA_harmony, reduction = "harmony", dims = 1:15) %>% FindClusters(resolution = 0.5)
Computing nearest neighbor graph
Computing SNN
Modularity Optimizer version 1.3.0 by Ludo Waltman and Nees Jan van Eck
Number of nodes: 14760
Number of edges: 529809
Running Louvain algorithm...
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Maximum modularity in 10 random starts: 0.9244
Number of communities: 12
Elapsed time: 3 seconds
降维可视化
scRNA_harmony <- RunUMAP(scRNA_harmony, reduction = "harmony", dims = 1:15)
11:41:02 UMAP embedding parameters a = 0.9922 b = 1.112
11:41:02 Read 14760 rows and found 15 numeric columns
11:41:02 Using Annoy for neighbor search, n_neighbors = 30
11:41:02 Building Annoy index with metric = cosine, n_trees = 50
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
11:41:09 Writing NN index file to temp file C:\Users\ASUS\AppData\Local\Temp\RtmpO8LMz0\file2da4113d7aa5
11:41:09 Searching Annoy index using 1 thread, search_k = 3000
11:41:16 Annoy recall = 100%
11:41:39 Commencing smooth kNN distance calibration using 1 thread
11:41:45 Initializing from normalized Laplacian + noise
11:41:47 Commencing optimization for 200 epochs, with 640848 positive edges
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
11:42:07 Optimization finished
画图看看整合效果
#group_by_cluster
plot1 = DimPlot(scRNA_harmony, reduction = "umap", label=T)
#group_by_sample
plot2 = DimPlot(scRNA_harmony, reduction = "umap", group.by='orig.ident')
#combinate
plotc <- plot1+plot2
plotc
对比一下Seurat整合样本，Seurat是采用锚定的方法来整合。
数据归一化处理
scRNA.list <- list(scRNA.2,scRNA.3)
for (i in 1:length(scRNA.list)) {
scRNA.list[[i]] <- NormalizeData(scRNA.list[[i]])
scRNA.list[[i]] <- FindVariableFeatures(scRNA.list[[i]], selection.method = "vst")
}
Performing log-normalization
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Calculating gene variances
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Calculating feature variances of standardized and clipped values
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Performing log-normalization
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Calculating gene variances
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Calculating feature variances of standardized and clipped values
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
寻找锚点
system.time({scRNA.anchors <- FindIntegrationAnchors(object.list = scRNA.list)})
Computing 2000 integration features
Scaling features for provided objects
|++++++++++++++++++++++++++++++++++++++++++++++++++| 100% elapsed=13s
Finding all pairwise anchors
| | 0 % ~calculating Running CCA
Merging objects
Finding neighborhoods
Finding anchors
Found 12929 anchors
Filtering anchors
Retained 4266 anchors
|++++++++++++++++++++++++++++++++++++++++++++++++++| 100% elapsed=15m 55s
user system elapsed
365.04 207.64 973.92
#锚定后整合
system.time({scRNA_seurat <- IntegrateData(anchorset = scRNA.anchors)})
scRNA_seurat<- IntegrateData(anchorset = scRNA.anchors)
Merging dataset 1 into 2
Extracting anchors for merged samples
Finding integration vectors
Finding integration vector weights
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Integrating data
save(scRNA.anchors,file='scRNA.anchors.RData')
rm(list = ls())
降维聚类
scRNA_seurat <- ScaleData(scRNA_seurat) %>% RunPCA(verbose=FALSE)
Centering and scaling data matrix
|===========================================================| 100%
scRNA_seurat <- FindNeighbors(scRNA_seurat, dims = 1:15) %>% FindClusters(resolution = 0.5)
Computing nearest neighbor graph
Computing SNN
Modularity Optimizer version 1.3.0 by Ludo Waltman and Nees Jan van Eck
Number of nodes: 14760
Number of edges: 523184
Running Louvain algorithm...
0% 10 20 30 40 50 60 70 80 90 100%
[----|----|----|----|----|----|----|----|----|----|
**************************************************|
Maximum modularity in 10 random starts: 0.9068
Number of communities: 12
Elapsed time: 3 seconds
scRNA_seurat <- ScaleData(scRNA_seurat, features = VariableFeatures(scRNA_seurat))
scRNA_seurat <- RunPCA(scRNA_seurat, features = VariableFeatures(scRNA_seurat))
plot1 <- DimPlot(scRNA_seurat, reduction = "pca", group.by="orig.ident")
plot2 <- ElbowPlot(scRNA_seurat, ndims=30, reduction="pca")
plotc <- plot1+plot2
pc.num=1:20
scRNA_seurat <- FindNeighbors(scRNA_seurat, dims = pc.num)
scRNA_seurat <- FindClusters(scRNA_seurat, resolution = 0.5)
table(scRNA_seurat@meta.data$seurat_clusters)
metadata <- scRNA_seurat@meta.data
cell_cluster <- data.frame(cell_ID=rownames(metadata), cluster_ID=metadata$seurat_clusters)
#tSNE
scRNA_seurat = RunTSNE(scRNA_seurat, dims = pc.num)
#group_by_cluster
plot1 = DimPlot(scRNA_seurat, reduction = "tsne", label=T)
#group_by_sample
plot2 = DimPlot(scRNA_seurat, reduction = "tsne", group.by='orig.ident')
#combinate
plotc <- plot1+plot2
#UMAP
scRNA_seurat <- RunUMAP(scRNA_seurat, dims = pc.num)
#group_by_cluster
plot3 = DimPlot(scRNA_seurat, reduction = "umap", label=T)
#group_by_sample
plot4 = DimPlot(scRNA_seurat, reduction = "umap", group.by='orig.ident')
#combinate
plotc <- plot3+plot4
plotc <- plot3+plot4+ plot_layout(guides = 'collect')
哈哈 暂时还看不出harmony与Seurat谁的整合效果好。。。
