Pairwise comparisons across species are problematic when analyzing functional genomic data

Pairwise (A, C, and E) and phylogenetic (B, D, and F) analyses of the original data (A and B), data simulated under the null hypothesis (C and D), and data simulated under the ortholog conjecture (E and F). In the pairwise plots, each point indicates the correlation coefficient of tau for a set of pairwise comparisons annotated with a specific node name (e.g., Primates) and event type (speciation or duplication, giving rise to orthologs and paralogs, respectively). The phylogenetic plots show the difference between the density distributions for tau phylogenetic contrasts for speciation and duplication events, where a value above zero indicates an excess of speciation contrasts in the indicated interval. A horizontal line at zero would indicate that the density distributions are identical. A reproduces the pattern presented in figure 2A of the work by KMRR (16) of higher correlation across speciation events than duplication events, which they took as evidence of the ortholog conjecture. The recovery of a similar pattern under both simulations (C and E) indicates that this pairwise approach does not make distinct testable predictions. The phylogenetic analysis of the original data (B) does not show an excess of larger contrasts for duplication events and does not reject the null hypothesis, providing no support for the ortholog conjecture. D and F validate the phylogenetic approach by showing that it does not reject the null when data are simulated under the null (D) but does reject the null when data are simulated under the ortholog conjecture (F). OC, ortholog conjecture.