Compute Quantitative Quality Metrics for Low-Dimensional Cell Embeddings

Computes quantitative cluster separability (Silhouette width), unsupervised cluster recovery (Adjusted Rand Index), and distribution preservation metrics (Mean UMI, Mean detected features, sparsity percentage, and relative percentage error) between reference and simulated datasets.

Usage

compute_embedding_quality_metrics(
  embedding_data,
  metric_coords = c("Dim1", "Dim2")
)

Arguments

embedding_data

A data.frame produced by compute_dataset_embeddings.

metric_coords

Character vector of the 2 coordinate columns used to compute distances (default: c("Dim1", "Dim2")).

Value

A data.frame summarizing quantitative embedding quality metrics for each dataset.

Examples

data(example_scrna, package = "scSimEval")
emb <- compute_dataset_embeddings(
  reference = example_scrna$ref,
  simulated = list("Splatter" = example_scrna$sim),
  reduction = "umap",
  cell_types = example_scrna$cell_types
)
#> Warning: You're computing too large a percentage of total singular values, use a standard svd instead.
#> Warning: You're computing too large a percentage of total singular values, use a standard svd instead.
compute_embedding_quality_metrics(emb)
#>     Dataset      Role Cells (N) Mean Silhouette ARI (Cluster Fidelity)
#> 1 Reference Reference        80          0.0009                 0.0277
#> 2  Splatter Simulated        80          0.0074                 0.0188
#>   Mean Library Size Mean Detected Features Library Size Diff (%)
#> 1             256.3                   49.2                  0.00
#> 2             242.3                   48.3                  5.46