Table 1 Algorithm 1 Top-down classification algorithm

      d, a new document

      G, a GO graph with trained base-classifiers

2

initialize:

3

   PA ← {root} //Predicted GO annotation set

4

   Q ← {root} //Queue for breadth first search

5

while Q not empty

6

   n ← Q.pop()

7

   S ← G.children(n)

8

   for each c in S:

9

      y ← c.predict (d)

         //Classify d using the base-classifier of c

10

      if y == 1:

            //if prediction is positive

11

         Q ← union(Q, c)

12

         PA ← union(PA, c)

13

   end

14

end

15

outputs: PA