Recovery of GDDIs [6].

<p>GDDIs make up 0.44% of DDIs in <b>A</b>, 0.60% in <b>B</b>, 0.21% in <b>C</b> and 0.11% in <b>D</b>. Median promiscuity of GDDIs to DDIs is 7.0∶5.0 in <b>A</b>, 8.5∶5.5 in <b>B</b>, 9.5∶5.5 in <b>C</b> and 15.0∶4.0 in <b>D</b>. Except for the concocted scenario <b>D</b>, at high Specificity (FPR ≤ 0.2), concept-based rankings produced with concept lattices that are not attribute-reduced (<b>OA</b> and <b>oA</b>) outperform (have higher Sensitivity or larger TPR) those by concept lattices that are attribute-reduced (<b>Oa</b> and <b>oa</b>). In scenarios <b>A</b> to <b>C</b>, <b>oA</b> outperforms <b>OA</b>. This outcome holds even when DOMINE <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0088943#pone.0088943-Yellaboina1" target="_blank">[18]</a> and 3did <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0088943#pone.0088943-Mosca1" target="_blank">[19]</a> domain-pairs are used to evaluate the rankings (<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0088943#pone-0088943-g016" target="_blank">Figure 16</a>). The quality of the <b>oA</b> domain-pair rankings becomes more evident by examining its top 100 domain-pairs (<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0088943#pone-0088943-g017" target="_blank">Figure 17</a> and <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0088943#pone.0088943.s002" target="_blank">File S2</a>). <b>OA</b>’s poorer performance in scenarios <b>A</b> to <b>C</b> is attributed to “excessive” piggy-backing. Scenario <b>D</b> shows that <b>oA</b>’s GDDI recovery is more sensitive to changes in protein domain architecture than <b>OA</b>’s. The changes introduced by domain shuffling in <b>D</b> reduce <b>oA</b>’s piggy-backing potential and its GDDI recovery suffers as a result. In contrast, <b>OA</b> is able to retain some of its previously “excessive” piggy-backing potential (<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0088943#pone-0088943-g014" target="_blank">Figure 14</a>). Nonetheless, the GDDIs recovered at low FPR by <b>OA</b> in scenario <b>D</b> are of the less promiscuous variety (<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0088943#pone-0088943-g012" target="_blank">Figure 12</a>). In all four scenarios, the <b>oa</b> concept lattice leaves at least 72% domain-pairs and at least 55.30% GDDIs scoreless. Also, the <b>Oa</b> rankings produced the worst GDDI recovery performance in all four scenarios.