scRNAseq clustering and visualization with path-metrics

Here we provide source code for a clustering and visualization method for single-cell RNA-seq data sets, that exploits path metrics. Using this method, distances between cells are measured in a data-driven way which is both density sensitive (decreasing distances across high density regions) and respects the underlying data geometry. By combining path metrics with multidimensional scaling, a low dimensional embedding of the data is obtained which respects both the global geometry of the data and preserves cluster structure.

Requirements

1. Adjusted Rand index Matlab function version 1.0.0.0
2. Munkres Assignment Algorithm Matlab function
3. Uniform Manifold Approximation and Projection (UMAP) Matlab function >= version 1.5.2
4. Matlab version 2021a

Description of files

• To reproduce results of processed data sets:

  1. Use the link in Processed data sets folder to access download data sets.
  2. Run RunPathMetrics.m using Matlab version 2021a.

• To produce the simulated beta cells data set:

  1. Download "GSM2230757_human1_umifm_counts.csv.gz" from here
  2. Run Simulationscript.R

• To process the data sets used in results:

  1. Download the real data sets using the links mentioned on “processing and imputation of data.R”
  2. Run processing and imputation of data.R
  3. Scale data
  • For Basic scaling: input in Matlab the csv file of the data set of interest produced in step 2, along with the true labels file. Then run Code/ProcessData.m after uncommenting the name of the data set and the line “Normalization = 'Basic';”.
  • For Linnorm scaling: run Linnorm transformation.R for the data set of interest produced in a. Then input in Matlab the output csv file along with the true labels file. Then run Code/ProcessData.m after uncommenting the name of the data set and the line “Normalization = 'Linnorm';”. Now you can use the output to run Code and data sets/Code_for_github/RunPathMetrics.m.
  • For SCT scaling: run sctranform_for_PM_after_SAVER_imputation.R for the data set of interest produced in a. Then input in Matlab the output csv file along with the true labels file. Then run Code/ProcessData.m after uncommenting the name of the data set and the line “Normalization = 'SCT';”. Now you can use the output to run Code/RunPathMetrics.m.

• Other files description:

  1. Seurat_clustering_analysis.R : Apply Seurat clustering on data set produced by processing and imputation of data.R
  2. Seurat_clustering_analysis_default_res: Apply Seurat clustering with resolution 0.8 on data set produced by processing and imputation of data.R
  3. RunSNNClustering.R: Apply SNN clustering on data sets after Basic or SCT scaling (created as described in 3b.)
  4. clustering_evaluation.R: Function that produces ARI, Entropy of cluster accuracy and Entropy of cluster Purity for a clustering.
  5. numerical_to_word_labels.xlsx: For interpretation purposed we provide the correspondence of numerical labels and descriptive labels for the data sets used for the results.