秋分:昼夜平分。
载入数据
此小插图演示了如何RunPCA
在Seurat v3.0中存储降维信息(例如的输出)并与之交互。出于演示目的,我们将使用在第一个指导教程中创建的2700 PBMC对象。您可以在此处下载预先计算的对象。
library(Seurat)
pbmc <- readRDS(file = "../data/pbmc3k_final.rds")
探索新的降维结构
在Seurat v3.0中,已将尺寸缩减信息的存储和交互作用概括化并形式化为DimReduc
对象。每个降维过程都作为DimReduc
对象存储在object@reductions
插槽中,作为命名列表的元素。可以通过操作[[
员调用所需的减少量的名称来访问这些减少量。例如,使用进行主成分分析后RunPCA
,object[['pca']]
将包含PCA的结果。通过将新元素添加到列表中,用户可以添加附加的和自定义的尺寸缩减。每个存储的尺寸缩减包含以下插槽:
- cell.embeddings:在低维空间中存储每个单元的坐标。
- feature.loadings:沿着嵌入的每个维度存储每个要素的权重
- feature.loadings.projected:Seurat通常会计算基因的子集(例如,高变异基因)的降维,然后将该结构投影到整个数据集(所有基因)上。该投影的结果(用计算
ProjectDim
)存储在此插槽中。请注意,投影后像元的加载将保持不变,但是现在所有要素都有要素加载 - stdev:每个尺寸的标准偏差。最常与PCA一起使用(存储协方差矩阵特征值的平方根),并且在查看由每个连续维数解释的方差量下降时很有用。
- key:设置cell.embeddings和feature.loadings矩阵的列名。例如,对于PCA,列名称为PC1,PC2等,因此键为“ PC”。
- jackstraw:存储使用此降维技术运行的jackstraw过程的结果。当前仅支持PCA。
- misc:额外的插槽,用于存储您可能需要的任何其他信息
要访问这些插槽,我们提供的Embeddings
,Loadings
和Stdev
功能
pbmc[["pca"]]
## A dimensional reduction object with key PC_
## Number of dimensions: 50
## Projected dimensional reduction calculated: FALSE
## Jackstraw run: TRUE
## Computed using assay: RNA
head(Embeddings(pbmc, reduction = "pca")[, 1:5])
## PC_1 PC_2 PC_3 PC_4 PC_5
## AAACATACAACCAC -4.7296855 -0.5184265 -0.7623220 -2.3156790 -0.07160006
## AAACATTGAGCTAC -0.5174029 4.5918957 5.9091921 6.9118856 -1.96243034
## AAACATTGATCAGC -3.1891063 -3.4695154 -0.8313710 -2.0019985 -5.10442765
## AAACCGTGCTTCCG 12.7933021 0.1007166 0.6310221 -0.3687338 0.21838204
## AAACCGTGTATGCG -3.1288078 -6.3481412 1.2507776 3.0191026 7.84739502
## AAACGCACTGGTAC -3.1088963 0.9262125 -0.6482331 -2.3244378 -2.00526763
head(Loadings(pbmc, reduction = "pca")[, 1:5])
## PC_1 PC_2 PC_3 PC_4 PC_5
## PPBP 0.010990202 0.01148426 -0.15176092 0.10403737 0.003299077
## LYZ 0.116231706 0.01472515 -0.01280613 -0.04414540 0.049906881
## S100A9 0.115414362 0.01895146 -0.02368853 -0.05787777 0.085382309
## IGLL5 -0.007987473 0.05454239 0.04901533 0.06694722 0.004603231
## GNLY -0.015238762 -0.13375626 0.04101340 0.06912322 0.104558611
## FTL 0.118292572 0.01871142 -0.00984755 -0.01555269 0.038743505
head(Stdev(pbmc, reduction = "pca"))
## [1] 7.098420 4.495493 3.872592 3.748859 3.171755 2.545292
Seurat提供了RunPCA
(pca)和RunTSNE
(tsne),并代表了通常应用于scRNA-seq数据的降维技术。使用这些功能时,所有插槽都会自动填充。
我们还允许用户添加单独计算的自定义降维技术(例如,多维缩放(MDS)或零膨胀因子分析)的结果。您需要的是一个矩阵,其中每个单元的坐标在低维空间中,如下所示。
存储自定义尺寸缩减计算
尽管未包含在Seurat软件包中,但它易于在R中运行多维缩放(MDS)。如果您对运行MDS并将输出存储在Seurat对象中感兴趣,请执行以下操作:
# Before running MDS, we first calculate a distance matrix between all pairs of cells. Here we
# use a simple euclidean distance metric on all genes, using scale.data as input
d <- dist(t(GetAssayData(pbmc, slot = "scale.data")))
# Run the MDS procedure, k determines the number of dimensions
mds <- cmdscale(d = d, k = 2)
# cmdscale returns the cell embeddings, we first label the columns to ensure downstream
# consistency
colnames(mds) <- paste0("MDS_", 1:2)
# We will now store this as a custom dimensional reduction called 'mds'
pbmc[["mds"]] <- CreateDimReducObject(embeddings = mds, key = "MDS_", assay = DefaultAssay(pbmc))
# We can now use this as you would any other dimensional reduction in all downstream functions
DimPlot(pbmc, reduction = "mds", pt.size = 0.5)
image.png
# If you wold like to observe genes that are strongly correlated with the first MDS coordinate
pbmc <- ProjectDim(pbmc, reduction = "mds")
## MDS_ 1
## Positive: CST3, TYROBP, FCER1G, LST1, FTL, AIF1, FTH1, TYMP, FCN1, LYZ
## LGALS1, S100A9, CFD, CD68, SERPINA1, CTSS, IFITM3, SPI1, S100A8, LGALS2
## Negative: MALAT1, RPS27A, RPS27, RPL3, RPL23A, RPL21, RPL13A, RPS6, RPS3A, RPS3
## RPL9, LTB, RPSA, CD3D, RPS25, RPS18, PTPRCAP, RPS12, RPL30, RPL31
## MDS_ 2
## Positive: NKG7, PRF1, CST7, GZMA, GZMB, B2M, FGFBP2, CTSW, GNLY, HLA-C
## GZMH, SPON2, CD247, FCGR3A, CCL5, HLA-A, CCL4, GZMM, KLRD1, CLIC3
## Negative: RPL32, RPL18A, HLA-DRA, CD79A, RPL13, MS4A1, RPL11, TCL1A, RPS9, RPL12
## LINC00926, HLA-DQB1, HLA-DQA1, HLA-DRB1, RPL28, RPS2, S100A8, HLA-DMA, RPL8, RPLP1
# Display the results as a heatmap
DimHeatmap(pbmc, reduction = "mds", dims = 1, cells = 500, projected = TRUE, balanced = TRUE)
image.png
# Explore how the first MDS dimension is distributed across clusters
VlnPlot(pbmc, features = "MDS_1")
image.png
# See how the first MDS dimension is correlated with the first PC dimension
FeatureScatter(pbmc, feature1 = "MDS_1", feature2 = "PC_1")
image.png