Skip to contents

Compute quantitative group-separation metrics per normalization method

Source: R/Normalization_Metrics.R
nm_compute_metrics.Rd

Iterates over assays in a SummarizedExperiment and computes ten metrics that quantify how well the normalization separates sample groups.

Usage

nm_compute_metrics(
  se,
  assay_names = NULL,
  condition_col = "Condition",
  seed = 42L,
  ...
)

Arguments

se

SummarizedExperiment from import_norm_matrices().

assay_names

Character vector of assay names to include. NULL = all.

condition_col

Column in colData with condition labels.

seed

Integer. Seed for the two metrics that use randomness: the Hopkins statistic, which samples random reference points, and the PERMANOVA p-value, which permutes the group labels. Without it neither column is reproducible between runs. The caller's RNG state is restored on exit. Default 42.

...

Additional arguments (unused).

Value

A data.frame with one row per method and columns: Method, PC1_VarPct, PC1_F_ratio, PERMANOVA_R2, PERMANOVA_pval, Silhouette_mean, MDS_GOF, MDS_CophCor, Spectral_Entropy, CumVar_PC2, Hopkins, Condition_Number.

Details

  • PC1_VarPct: % variance explained by PC1 (higher = more structure).

  • PC1_F_ratio: ANOVA F-ratio on PC1 scores (higher = better separation).

  • PERMANOVA_R2: Proportion of variance explained by grouping (requires vegan; NA if not installed).

  • Silhouette_mean: Mean silhouette width (requires cluster; NA if not installed).

  • MDS_GOF: Goodness-of-fit of 2D MDS projection. Note: in benchmarking context, lower values tend to indicate better normalization (a good method preserves multi-dimensional biological variability that 2D cannot capture).

  • MDS_CophCor: Cophenetic correlation between original and MDS distances. Same caveat as MDS_GOF: lower values tend to correlate with better benchmarking performance.

  • Spectral_Entropy: Normalized Shannon entropy of PCA eigenvalues (0-1). Values near 0 = variance concentrated in few PCs; near 1 = uniform spread.

  • CumVar_PC2: Cumulative % variance explained by PC1 + PC2 (0-100).

  • Hopkins: Hopkins statistic for clustering tendency (0-1). Values > 0.5 suggest non-random cluster structure.

  • Condition_Number: Ratio of largest to smallest PCA eigenvalue (>= 1). High values indicate multicollinearity or numerical instability.

Examples

data(nadia_dia)
se <- nm_prepare_se(nadia_dia, verbose = FALSE)
se <- nm_run_normalizations(se, methods = c("cycloess", "quantile", "MAD"),
                            verbose = FALSE)

metrics_df <- nm_compute_metrics(se)
metrics_df[, c("Method", "PC1_VarPct", "PC1_F_ratio", "Silhouette_mean")]
#>     Method PC1_VarPct PC1_F_ratio Silhouette_mean
#> 1     log2   56.03859    120.9196       0.2118819
#> 2 cycloess   72.57836   9790.2845       0.3790455
#> 3 quantile   65.48778  10168.4649       0.3760405
#> 4      MAD   64.86712   9803.3024       0.3862026