Adjust ontology levels figure

[bschilder]

• See what p-value vs. ontology level looks like
• Remove specificity.
• Add labels to the y-axis to make it more obvious what ontology level means (more broad terms <--> more specific terms)

[bschilder]

p-values (all)

When you plot all phenotype-celltype pvalues against ontology, this really just reflects the number of significant associations. This shows what we expect, fewer significant pvalues with more specific phenotypes.

p-values (significant only)

If you subset to tests to only those that passed the FDR<0.05 threshold and then plot the pvalues, there's a very significant (but very small) effect of more significant pvalues with more specific phenotypes.

[bschilder]

preview of the arrows annotating the y-axis

[bschilder]

When I checked whether logging the pvalues and plotting only the sig ones, I don't think it made any difference. But i'll go back and try again manually to make sure.

[bschilder]

Would also be worth plotting against Information Content, which is a metric that approximates ontological terms specificity while normalising for different branch depths.

[bschilder]

results <- MSTExplorer::load_example_results()
results <- HPOExplorer::add_hpo_name(results, hpo = hpo)
results <- HPOExplorer::add_ont_lvl(results)

Ontology level vs. Genes, Cell Types (sig), and P-values

plot_ontology_levels_out <- MSTExplorer::plot_ontology_levels(
  results = results, 
  ctd_list = ctd_list,
  x_vars = c("genes","cell types","p"), 
  nrow = 1 ) 

Ontology level vs. Genes, Cell Types (sig), and P-values (sig)

plot_ontology_levels_out <- MSTExplorer::plot_ontology_levels(
  results = results, 
  ctd_list = ctd_list,
  x_vars = c("genes","cell types","p"),
  sig_vars= c(FALSE, TRUE, TRUE),
  log_vars = c(FALSE, FALSE, FALSE),
  nrow = 1) 

Ontology level vs. Genes, Cell Types (sig), and P-values (logged)

When logging p-values, I replace p-values of exactly 0 to avoid resulting in Inf.
I replace all p-values==0 with the smallest number R can compute:

> .Machine$double.xmin
[1] 2.225074e-308

plot_ontology_levels_out <- MSTExplorer::plot_ontology_levels(
  results = results, 
  ctd_list = ctd_list,
  x_vars = c("genes","cell types","p"),
  sig_vars= c(FALSE, TRUE, FALSE),
  log_vars = c(FALSE, FALSE, TRUE),
  nrow = 1) 

Ontology level vs. Genes, Cell Types (sig), and P-values (sig, logged)

plot_ontology_levels_out <- MSTExplorer::plot_ontology_levels(
  results = results, 
  ctd_list = ctd_list,
  x_vars = c("genes","cell types","p"),
  sig_vars= c(FALSE, TRUE, TRUE),
  log_vars = c(FALSE, FALSE, TRUE),
  nrow = 1) 

Conclusion

Plotting all non-logged p-values seems to be the most interpretable to me.