Figure 5.

Optimization of DCA. Only pairs separated by at least five positions in primary sequence are considered. (A–E) DCA was applied to FhaC as in Figure 1, A and B, with the same definition of true positives. (A) Effect of shrinkage intensity α on DCA true positive rates. (B) Effect of shrinkage intensity α = 0.6 on DI_ij scores. Fold reduction in DI_ij score is plotted against the minimum pair entropy S_min(i,j) for each pair; the red curve shows the median fold reduction in DI_ij score over 50 bins of S_min(i,j). (C and D) DI_ij scores for shrinkage intensity α = 0.6 plotted against minimum pair entropy S_min(i,j). The top 50 pairs according to (C) DI_ij and (D) DIZ_ij scoring (α = 0.6) are highlighted. In C the dashed black line is the cutoff for pairs with the 50 highest DI_ij scores. (E and F) Overlap of the top 50 (E) FhaC and (F) BamA pairs according to DI_ij scoring (α = 0) and DIZ_ij scoring (α = 0.6). Horizontal and vertical dashed lines correspond to the cutoffs for pairs with the 50 highest DI_ij and DIZ_ij scores, respectively. (F) DCA was applied to BamA as in Figure 1D, with the same definition of true positives. There are no false positives. Arrow indicates pair R661–D740.