Fig. 1.

The brief procedure of the annealed importance sampling. (a) Protein interaction network of a gene set. It is not necessary that this network has to be a connected graph. (b) The procedure sampling N _k from p(N). This annealed run generates a sequence of networks: n _m − 1, n _m − 2, …, n ₀, which are drawn from p _j kept by temperature T _j for the annealing run and genetic interaction score (j = 0,  … , m − 1). Where,$p_j=p^{1-{\beta }_j}p{\left(N\right)}^{{\beta }_j}$, 1 = β ₀ > β ₁ >  …  > β _m − 1 = 0, the p is a uniform distribution from which an initial network n _m − 1 can be drawn easily. Then let N _k = n ₀, and save the weight ω _j. (c) Genetic interaction profiles. Using the profiles, score of candidate network can be computed, and the distribution p(N), which is proportional f(n) shown as eq. (2), can be defined