Table 3.

GeneId-Ranking With a Graph

(1) let Q0 be the probability distribution such that Q0(x) = 1

(2) for d = 1, ..., dmax do

• let Qd(x) = 0 for all x

• for i = 1, ..., m do

- sample xi according to Qd-1

- Qd(xi) = γQd-1 (xi)

- for each edge label ℓ ∈ L (x)

* for each node y ∈ Y(x, ℓ)

· let qxy = Pr(y|ℓ, x)·Pr(ℓ|x)

· increment Qd(y) by (1 - γ)Qd-1 (xi)qxy

(3) return $Q_{d_{max}}$ (z) as an approximation to Q(z|x)

An efficient approximation algorithm for computing Q(z|x), given transition probabilities Pr(y|x, ℓ) and Pr(ℓ|x).