Statistical Analysis • CellTFusion

library(CellTFusion)
#> 
#>

Once cell groups have been constructed (see Cell Group Construction), you can evaluate their association with clinical traits using a range of statistical tests.

Statistical association tests

The unified function scores.stat.analysis() accepts:

Cell group scores from construct_cell_groups()
NMF latent factors from compute.latent_factors()
Any samples × features numeric matrix

Supported methods are "fisher", "anova", "wilcox", "kruskal", and "ttest".

# Fisher's exact test
res_fisher <- scores.stat.analysis(
  scores  = cell_groups,
  coldata = traitdata,
  trait   = "Best.Confirmed.Overall.Response",
  method  = "fisher",
  pval    = 0.05
)

# ANOVA
res_anova <- scores.stat.analysis(
  scores  = cell_groups,
  coldata = traitdata,
  trait   = "Best.Confirmed.Overall.Response",
  method  = "anova",
  pval    = 0.05
)

# Wilcoxon rank-sum test
res_wilcox <- scores.stat.analysis(
  scores  = cell_groups,
  coldata = traitdata,
  trait   = "Best.Confirmed.Overall.Response",
  method  = "wilcox",
  pval    = 0.05
)

# Kruskal–Wallis test
res_kruskal <- scores.stat.analysis(
  scores  = cell_groups,
  coldata = traitdata,
  trait   = "Best.Confirmed.Overall.Response",
  method  = "kruskal",
  pval    = 0.05
)

Latent factor analysis

compute.latent_factors() decomposes cell group scores into NMF latent factors, which can then be passed directly to scores.stat.analysis():

nmf <- compute.latent_factors(cell_groups)

res_nmf <- scores.stat.analysis(
  scores  = nmf,
  coldata = traitdata,
  trait   = "Best.Confirmed.Overall.Response",
  method  = "anova",
  pval    = 0.05
)

Clinical metadata association

To explore associations between TF module scores and all available clinical traits simultaneously, use compute.metadata.association(). It applies Pearson correlation for continuous traits and ANOVA for categorical traits, and saves a labeled heatmap and violin plots to Results/:

compute.metadata.association(
  tfs.modules = network[[1]],
  coldata     = traitdata,
  pval        = 0.05,
  file.name   = "Tutorial",
  width       = 10
)