In this tutorial we will continue the analysis of the integrated dataset. We will use the integrated PCA to perform the clustering. First we will construct a \(k\)-nearest neighbour graph in order to perform a clustering on the graph. We will also show how to perform hierarchical clustering and k-means clustering on PCA space.

Let’s first load all necessary libraries and also the integrated dataset from the previous step.


alldata <- readRDS("data/results/covid_qc_dr_int.rds")

Graph clustering

The procedure of clustering on a Graph can be generalized as 3 main steps:

  1. Build a kNN graph from the data

  2. Prune spurious connections from kNN graph (optional step). This is a SNN graph.

  3. Find groups of cells that maximizes the connections within the group compared other groups.

Building kNN / SNN graph

The first step into graph clustering is to construct a k-nn graph, in case you don’t have one. For this, we will use the PCA space. Thus, as done for dimensionality reduction, we will use ony the top N PCA dimensions for this purpose (the same used for computing UMAP / tSNE).

As we can see above, the Seurat function FindNeighbors already computes both the KNN and SNN graphs, in which we can control the minimal percentage of shared neighbours to be kept. See ?FindNeighbors for additional options.

# check that CCA is still the active assay

alldata <- FindNeighbors(alldata, dims = 1:30, k.param = 60, prune.SNN = 1/15)
## Computing nearest neighbor graph
## Computing SNN
# check the names for graphs in the object.
## [1] "CCA"
## [1] "CCA_nn"  "CCA_snn"

We can take a look at the kNN graph. It is a matrix where every connection between cells is represented as \(1\)s. This is called a unweighted graph (default in Seurat). Some cell connections can however have more importance than others, in that case the scale of the graph from \(0\) to a maximum distance. Usually, the smaller the distance, the closer two points are, and stronger is their connection. This is called a weighted graph. Both weighted and unweighted graphs are suitable for clustering, but clustering on unweighted graphs is faster for large datasets (> 100k cells).

pheatmap(alldata@graphs$CCA_nn[1:200, 1:200], col = c("white", "black"), border_color = "grey90",
    main = "KNN graph", legend = F, cluster_rows = F, cluster_cols = F, fontsize = 2)

pheatmap(alldata@graphs$CCA_snn[1:200, 1:200], col = colorRampPalette(c("white",
    "yellow", "red"))(100), border_color = "grey90", main = "SNN graph", legend = F,
    cluster_rows = F, cluster_cols = F, fontsize = 2)

Clustering on a graph

Once the graph is built, we can now perform graph clustering. The clustering is done respective to a resolution which can be interpreted as how coarse you want your cluster to be. Higher resolution means higher number of clusters.

In Seurat, the function FindClusters will do a graph-based clustering using “Louvain” algorithim by default (algorithm = 1). TO use the leiden algorithm, you need to set it to algorithm = 4. See ?FindClusters for additional options.

# Clustering with louvain (algorithm 1)
for (res in c(0.1, 0.25, 0.5, 1, 1.5, 2)) {
    alldata <- FindClusters(alldata, = "CCA_snn", resolution = res, algorithm = 1)

# each time you run clustering, the data is stored in meta data columns:
# seurat_clusters - lastest results only CCA_snn_res.XX - for each different
# resolution you test.

plot_grid(ncol = 3, DimPlot(alldata, reduction = "umap", = "CCA_snn_res.0.5") +
    ggtitle("louvain_0.5"), DimPlot(alldata, reduction = "umap", = "CCA_snn_res.1") +
    ggtitle("louvain_1"), DimPlot(alldata, reduction = "umap", = "CCA_snn_res.2") +