Figure 4

Illustration of the graph pyramid search subroutine. Here we elaborate the algorithm 1 using 5 different queries. For simplicity assume T AC = 0, t AC = {t1, t2, t3} and only 3 protein classes i.e. . Each query q i , is analyzed first at the highest level (t3) where we look for all classes , having ΔAC(c, t3) > T AC and collect them in . For q1 we can not find any such classes c at t3, so we descend the GP to t2 level and discover with threshold t∗ = t2. This secondary threshold (T AC ) is necessary, otherwise there will be many spurious classes (false positives (FP)) having nonzero ΔAC(·).