Skip to content

Latest commit

 

History

History
44 lines (30 loc) · 1.76 KB

v3_0.md

File metadata and controls

44 lines (30 loc) · 1.76 KB

Overview

We expect a snn_graph_cluster HDF5 group at the root of the file, containing details on the SNN graph-based clustering. The group itself should contain the parameters and results subgroups.

Definitions:

  • in_use: whether SNN graph clustering is the chosen method in choose_clustering.
  • num_cells: number of cells remaining after QC filtering.

Parameters

parameters will contain:

  • k: a scalar integer specifying the number of nearest neighbors to find.
  • scheme: a scalar string specifying the edge weighting scheme to use. This may be "rank", "number" or "jaccard".
  • algorithm: a scalar string specifying the community detection method to use. This may be "multilevel", "walktrap" or "leiden".
  • multilevel_resolution: a scalar float specifying the resolution of the multi-level community detection. This should be non-negative.
  • leiden_resolution: a scalar float specifying the resolution of the leiden algorithm. This should be non-negative.
  • walktrap_steps: a scalar integer specifying the number of steps to use in the Walktrap algorithm. This should be non-negative.

Results

If in_use = true, results should contain:

  • clusters: an integer dataset of length equal to num_cells, containing the SNN graph cluster assignment for each cell. For N clusters, there should be at least one occurrence of each integer in [0, N).

If in_use = false, clusters may be absent. If it is present, it should follow the constraints listed above.

History

Updated in version 3.0, with the following changes from the previous version:

  • Added choice of community detection algorithm.
  • Added algorithm-specific parameters.
  • Renamed resolution to multilevel_resolution.