Clustering
Material
- Evaluation of clustering methods
Exercises
Load the seu_int dataset you have created earlier today:
seu_int <- readRDS("seu_int_day2_part1.rds")
The method implemented in Seurat first constructs a SNN graph based on the euclidean distance in PCA space, and refine the edge weights between any two cells based on the shared overlap in their local neighborhoods (Jaccard similarity). This step is performed using the FindNeighbors() function, and takes as input the previously defined dimensionality of the dataset.
Note
We use the integrated object (seu_int) and the assay integrated. Unsure? Check DefaultAssay(seu_int), and set it by DefaultAssay(seu_int) <- "integrated".
seu_int <- Seurat::FindNeighbors(seu_int, dims = 1:25)
To cluster the cells, Seurat next implements modularity optimization techniques such as the Louvain algorithm (default) or SLM [SLM, Blondel et al., Journal of Statistical Mechanics], to iteratively group cells together, with the goal of optimizing the standard modularity function. The FindClusters() function implements this procedure, and contains a resolution parameter that sets the ‘granularity’ of the downstream clustering, with increased values leading to a greater number of clusters.
seu_int <- Seurat::FindClusters(seu_int, resolution = seq(0.1, 0.8, by=0.1))
Cluster id of each cell is added to the metadata object, as a new column for each resolution tested:
head(seu_int@meta.data)
To view how clusters sub-divide at increasing resolution:
library(clustree)
clustree::clustree(seu_int@meta.data[,grep("integrated_snn_res", colnames(seu_int@meta.data))],
prefix = "integrated_snn_res.")
You can view the UMAP coloring each cell according to a cluster id like this:
Seurat::DimPlot(seu_int, group.by = "integrated_snn_res.0.1")
Exercise: Visualise clustering based on a few more resolutions. Taking the clustering and the UMAP plots into account what do you consider as a good resolution to perform the clustering?
Answer
Of course, there is no ‘optimal’ resolution, but based on resolution of 0.3, the tree stays relatively stable for a few resolution steps, and it seems that clustering fits the UMAP well:
Seurat::DimPlot(seu_int, group.by = "integrated_snn_res.0.3")
Exercise: When do the number of neighbors need to be changed? How does changing the method of clustering in FindClusters affect the output? Which parameter should be changed?
Answer
As FindClusters is an unsupervised clustering method on the PCA data and UMAP is a good summary of the PCA dimension selected, clusters and UMAP plot should go along. If one has reasons to change the number of neighbors in the UMAP function, here the same parameter should be adapted.
The method can be changed with algorithm = 2,3 or 4