Question: What types of dimensionality reduction and clustering are performed by Cell Ranger ATAC?
Answer: Cell Ranger ATAC performs dimensionality reduction via Principal Component Analysis (PCA), Latent Semantic Analysis (LSA), or Probabilistic Latent Semantic Analysis (PLSA). The adopted default method is LSA, but users can specify which method to use through the --dim-reduce parameter.
--dim-reduce |
(optional) Chose the algorithm for dimensionality reduction prior to clustering and tsne: 'lsa' (default), 'plsa', or 'pca'. |
After dimensionality reduction, graph-based clustering is always performed. In addition, for PCA, Cell Ranger ATAC performs k-means clustering that produces 2 to 10 clusters for visualization and analysis. If LSA or PLSA is used for dimensionality reduction, then Cell Ranger ATAC performs k-medoids clustering that produces 2 to 10 clusters for downstream analysis.
Below is the summary of dimensionality reduction techniques and associated clustering and visualization approaches provided by the pipeline.
| Dimensionality Reduction | Clustering |
|---|---|
| PCA | K-means, graph-based clustering |
| LSA | K-medoids, graph-based clustering |
| PLSA | K-medoids, graph-based clustering |
For more information on this topic, please consult the clustering documentation page.