Figure 2

An example illustrating the benefit of balanced profiles. Balanced profiles guarantee equal weight on the four scenarios of coevolution between genes, while raw profiles contain varying weights. A: The first two steps in CMP-KDO biosynthesis I in E. coli [20]. Enzymes gutQ and kdsD are both D-arabinose-5-phosphate isomerase; enzyme kdsA is 3-deoxy-D-manno-octulosonate-8-phosphate synthase, and PEP is phosphoenolpyruvate. The first step can be catalyzed by either gutQ or kdsD, and the second step by kdsA. The three genes form a type 3 logic relationship, "kdsA is present in a genome if and only if gutQ or kdsD is present". B: Using raw profiles, two significant triplets are identified with ΔU > 0.3, along with four other triplets with ΔU > 0.1. Neither of the two significant triplets is consistent with the network in E. coli or has experimental support, and the expected type 3 triplet only ranks third with a low insignificant ΔU. In comparison, the balanced profiles produce top score for the expected triplet, significantly lower the score of an unfitting triplet from 0.412 to -0.075, and identify only three triplets with ΔU > 0.1; that is, much less noise.