cluster
不同聚类方法比较
normmlization，去除一些批次效应等外部因素。通过对表达矩阵的矩阵的聚类，对细胞群体进行分类。
无监督聚类：hierarchical clustering，k-means clustering，graph-based clustering。
library(SingleCellExperiment)
deng <- readRDS("D:/paopaoR/mass/deng-reads.rds")
deng
## class: SingleCellExperiment
## dim: 22431 268
## metadata(0):
## assays(2): counts logcounts
## rownames(22431): Hvcn1 Gbp7 ... Sox5 Alg11
## rowData names(10): feature_symbol is_feature_control ...
## total_counts log10_total_counts
## colnames(268): 16cell 16cell.1 ... zy.2 zy.3
## colData names(30): cell_type2 cell_type1 ... pct_counts_ERCC
## is_cell_control
## reducedDimNames(0):
## spikeNames(1): ERCC
table(deng$cell_type2)
##
## 16cell 4cell 8cell early2cell earlyblast late2cell
## 50 14 37 8 43 10
## lateblast mid2cell midblast zy
## 30 12 60 4
table(deng$cell_type1)
##
## 16cell 2cell 4cell 8cell blast zygote
## 50 22 14 37 133 12
library(scater)
## Warning: package 'scater' was built under R version 3.5.2
plotPCA(deng, colour_by = "cell_type2")
## Warning: 'add_ticks' is deprecated.
## Use '+ geom_rug(...)' instead.
image.png
plotPCA(deng, colour_by = "cell_type1")
## Warning: 'add_ticks' is deprecated.
## Use '+ geom_rug(...)' instead.
image.png
SC3
当细胞数量不是很多的时候可以用sc3，共识聚类，可以一步分析到位，也可以根据需要分步分析。但当细胞数量很多的时候速度会很慢，用seurat会比较方便。
library(SC3)
deng <- sc3_estimate_k(deng)
metadata(deng)$sc3$k_estimation
## [1] 6
deng <- sc3(deng, ks = 10, biology = TRUE)
## Setting SC3 parameters...
## Calculating distances between the cells...
## Performing transformations and calculating eigenvectors...
## Performing k-means clustering...
## Calculating consensus matrix...
## Calculating biology...
colnames(rowData(deng))
## [1] "feature_symbol" "is_feature_control"
## [3] "is_feature_control_ERCC" "mean_counts"
## [5] "log10_mean_counts" "rank_counts"
## [7] "n_cells_counts" "pct_dropout_counts"
## [9] "total_counts" "log10_total_counts"
## [11] "sc3_gene_filter" "sc3_10_markers_clusts"
## [13] "sc3_10_markers_padj" "sc3_10_markers_auroc"
## [15] "sc3_10_de_padj"
colnames(colData(deng))
## [1] "cell_type2"
## [2] "cell_type1"
## [3] "total_features"
## [4] "log10_total_features"
## [5] "total_counts"
## [6] "log10_total_counts"
## [7] "pct_counts_top_50_features"
## [8] "pct_counts_top_100_features"
## [9] "pct_counts_top_200_features"
## [10] "pct_counts_top_500_features"
## [11] "total_features_endogenous"
## [12] "log10_total_features_endogenous"
## [13] "total_counts_endogenous"
## [14] "log10_total_counts_endogenous"
## [15] "pct_counts_endogenous"
## [16] "pct_counts_top_50_features_endogenous"
## [17] "pct_counts_top_100_features_endogenous"
## [18] "pct_counts_top_200_features_endogenous"
## [19] "pct_counts_top_500_features_endogenous"
## [20] "total_features_feature_control"
## [21] "log10_total_features_feature_control"
## [22] "total_counts_feature_control"
## [23] "log10_total_counts_feature_control"
## [24] "pct_counts_feature_control"
## [25] "total_features_ERCC"
## [26] "log10_total_features_ERCC"
## [27] "total_counts_ERCC"
## [28] "log10_total_counts_ERCC"
## [29] "pct_counts_ERCC"
## [30] "is_cell_control"
## [31] "sc3_10_clusters"
## [32] "sc3_10_log2_outlier_score"
sc3_plot_consensus(deng, k = 10, show_pdata = "cell_type2")
image.png
sc3_plot_silhouette(deng, k = 10)
image.png
sc3_plot_expression(deng, k = 10, show_pdata = "cell_type2")
image.png
sc3_plot_markers(deng, k = 10, show_pdata = "cell_type2")
image.png
plotPCA(deng, colour_by = "sc3_10_clusters")
## Warning: 'add_ticks' is deprecated.
## Use '+ geom_rug(...)' instead.
image.png
library(mclust)
adjustedRandIndex(colData(deng)$cell_type2, colData(deng)$sc3_10_clusters)
## [1] 0.6616899
pcaReduce
library(pcaReduce)
input <- logcounts(deng[rowData(deng)$sc3_gene_filter, ])
pca.ite <- PCAreduce(t(input), nbt = 1, q = 30, method = 'S')[[1]]#nbt-number of samples,q-number of dimensions to start with
colData(deng)$pcaReduce <- as.character(pca.ite[,32 - 10])#k-(2~q+1),k=10,(q+1)-10+1
plotPCA(deng, colour_by = "pcaReduce")
## Warning: 'add_ticks' is deprecated.
## Use '+ geom_rug(...)' instead.
image.png
adjustedRandIndex(colData(deng)$cell_type2, colData(deng)$pcaReduce)
## [1] 0.3504523
tsne+k-means
set.seed(1234)
deng <- runTSNE(deng,perplexity = 50)#change perplexity to have robust result
colData(deng)$tSNE_kmeans8 <- as.character(kmeans(deng@reducedDims$TSNE, centers = 8)$clust)
plotTSNE(deng,colour_by = "tSNE_kmeans8")
## Warning: 'add_ticks' is deprecated.
## Use '+ geom_rug(...)' instead.
image.png
colData(deng)$tSNE_kmeans10 <- as.character(kmeans(deng@reducedDims$TSNE, centers = 10)$clust)
plotTSNE(deng,colour_by = "tSNE_kmeans10")
## Warning: 'add_ticks' is deprecated.
## Use '+ geom_rug(...)' instead.
image.png
adjustedRandIndex(colData(deng)$cell_type2, colData(deng)$tSNE_kmeans10)
## [1] 0.3753199
deng <- runTSNE(deng,perplexity = 10)
colData(deng)$tSNE_kmeans10.1 <- as.character(kmeans(deng@reducedDims$TSNE, centers = 10)$clust)
adjustedRandIndex(colData(deng)$cell_type2, colData(deng)$tSNE_kmeans10.1)
## [1] 0.4531443
deng <- runTSNE(deng,perplexity = 5)
colData(deng)$tSNE_kmeans10.2 <- as.character(kmeans(deng@reducedDims$TSNE, centers = 10)$clust)
adjustedRandIndex(colData(deng)$cell_type2, colData(deng)$tSNE_kmeans10.2)
## [1] 0.48855
SINCERA
这个包感觉用的不是很多，感觉就是用了个zscore转换，根据相关层次聚类，迭代找个k值。
dat <- apply(input, 1, function(y) scRNA.seq.funcs::z.transform.helper(y))# perform z-score transformation
dd <- as.dist((1 - cor(t(dat), method = "pearson"))/2)# hierarchical clustering
hc <- hclust(dd, method = "average")
num.singleton <- 0
kk <- 1
for (i in 2:dim(dat)[2]) {
clusters <- cutree(hc, k = i)
clustersizes <- as.data.frame(table(clusters))
singleton.clusters <- which(clustersizes$Freq < 2)
if (length(singleton.clusters) <= num.singleton) {
kk <- i
} else {
break;
}
}
cat(kk)
## 6
library(pheatmap)
pheatmap(
t(dat),
cluster_cols = hc,
cutree_cols = kk,
kmeans_k = 100,
show_rownames = FALSE
)
image.png
adjustedRandIndex(colData(deng)$cell_type2,clusters)
## [1] 0.3827252
clusters1 <- cutree(hc, k = 10)
adjustedRandIndex(colData(deng)$cell_type2,clusters1)
## [1] 0.3748432
dynamicTreeCut
deng <- runPCA(deng)
my.dist <- dist(reducedDim(deng))
my.tree <- hclust(my.dist, method="ward.D2")
library(dynamicTreeCut)
my.clusters <- unname(cutreeDynamic(my.tree, distM=as.matrix(my.dist),
minClusterSize=1, verbose=0))#modify cutheight to change clusters
table(my.clusters)
## my.clusters
## 1 2 3 4 5 6 7 8
## 55 52 52 34 26 22 14 13
deng$clusters <- factor(my.clusters)
plotTSNE(deng,colour_by="clusters")
## Warning: 'add_ticks' is deprecated.
## Use '+ geom_rug(...)' instead.
image.png
adjustedRandIndex(colData(deng)$cell_type2,my.clusters)
## [1] 0.5354439
SNN clip
基于graph理论，找共有邻近点，现在的seurat也是基于graph，用knn，snn什么的，可以改变共享邻近点的数目改变cluster结果。
library(igraph)
library(scran)
snn.gr <- buildSNNGraph(deng, use.dimred="PCA")
cluster.out <- igraph::cluster_walktrap(snn.gr)
my.clusters <- cluster.out$membership
table(my.clusters)
## my.clusters
## 1 2 3 4 5 6 7 8 9
## 53 36 13 22 40 33 36 21 14
deng$cluster <- factor(my.clusters)
plotTSNE(deng, colour_by="cluster")
## Warning: 'add_ticks' is deprecated.
## Use '+ geom_rug(...)' instead.
image.png
set.seed(123)
reducedDim(deng, "force") <- igraph::layout_with_fr(snn.gr, niter=3000)
plotReducedDim(deng, colour_by="cluster", use_dimred="force")
## Warning: 'add_ticks' is deprecated.
## Use '+ geom_rug(...)' instead.
image.png
adjustedRandIndex(colData(deng)$cell_type2,my.clusters)
## [1] 0.4505625
RCA
据说是现有的聚类效果最好的包，用分析的数据投影一个参考来进行聚类，但是模式还不是很成熟，暂时还没有小鼠的模式，而且前期流程跑下来的数据是fpkm，也许有些影响。并且要求名字格式“XXXX_genenames_YYYY"，前面是位置信息，后面是ens号，emmm不知道为啥搞成这样，毕竟最后它还是提取中间的名字进行分析。。。。
RCA, short for Reference Component Analysis, is an R package for robust clustering analysis of single cell RNA sequencing data (scRNAseq). It is developed by Prabhakar Group at Genome Institute of Singapore (GIS).
Features: -clustering analysis of scRNAseq data from Human samples -three modes : “GlobalPanel”: default option for clusterig general single cell data sets that include a wide spectrum of cell types. “ColonEpitheliumPanel”: suitable for analyzing human colon/intestine derived samples. “SelfProjection”: suitable for analyzing data sets from not-well-studied tissue types (still under optimization)