Fig. 4

Clustering disease property tests by their functional convergence trends. a The correlations of the ranked effect size trends are significantly different from null distributions for the 61 functional properties (effect size permutation null, Student’s paired t-test p < 2.2e-16). We’ve drawn the red lines to indicate the FDRs of 0.01, where 14 functions are significant, and 0.05, where 44 are significant. b If we filter for the functions with some weak signal in the underlying functional tests (p < 0.5), 383 correlations are considered with ten functions as significant (dark blue, Student’s paired t-test p < 2.2e-16). Note that we are not filtering with respect to our own functional effect size test, which assesses variation in the underlying functional tests, merely that the underlying tests do return some values. c When we have no constraints (all tests, 4164 shown), three pass an FDR of 0.01. d We enumerate these in a barplot. e A heatmap of all the ranked scores of the test gene sets (columns) for the subset of 61 significant effect properties (rows). The properties clustered into six groups when we cut the dendrogram at a height of ~12. Their functional convergence correlations (ranked) show that most high correlations cluster (in clusters 1 and 3). White/yellow is a high rank, red is low. The property type is color coded as described in the figure key. A high correlation is shaded purple, low/negative correlations are grey. Clusters are labeled and colored, with function FDR <0.01 outlined in black