Figure 3

Connectivity trends in gene graphs built by each model. Analyzing the increase of connectivity coverage as the stringency level (coefficient cutoff) decreases. A and B, for every cutoff, the accepted portion of the graph is plotted as the percentage of all genes (solid line, left y-axis) and the number of gene pair connections (dashed line, right y-axis). All three graphs include >98% of the genes at the lowest plotted stringency level, though the dynamics of how the graphs grow are different, separating the RA method from PE and MI. Likewise in gene pair relationship inclusion, with the additional distinction that as the ratiometric model does only consider connections that exhibit a ratiometric profile it will have a smaller total number of possible connections (24% of total gene pairs possible). The dashed black vertical line marks the stringency level at which 95% of the genes are included, |V(G)| = 9244. B) The overlap of genes and gene pairs between the three methods at different vertex sizes, plotted against number of genes accepted (upper graph) or number of gene pairs accepted (lower graph). For each size of the graph, the following fractions of the total number of discoveries (genes or gene pairs) are given: RA, MI, PE, RA-MI, RA-PE, RA-MI-PE, MI-PE, where RA = ratiometric method, PE = squared Pearson correlation and MI = mutual information. PE and MI have a larger overlap compared to the RA method that differs compared to both former ones.