Clustering Cells in Single Cell Analysis
How do we cluster cells in single-cell RNA-seq?
Graph-based clustering (Seurat) is the most popular method.
FindNeighbors() has default k = 20, k-nearest neighbor (kNN), 20 cells
https://bioconductor.org/books/3.15/OSCA.basic/clustering.html
- Similarity is computed. Euclidean distance or cosine similarity
- Nearest neighbors are found
- A graph (network) is built. Cells = nodes, neighbor relationships = edges, stronger similarity = stronger edge weight.
- Communities (clusters) are detected. Clustering algorithms in single cell: Louvain or Leiden