Table 2.

Protein complex identification algorithm

Input

- Weighted PPI matrix;

- A training set of complexes and non-complexes;

Output

- Discovered list of protein complexes;

Complex model parameter estimation

- Extract property features from positive and negative training examples;

- Discretize the continuous features;

- Calculate the BN MLE parameters for different features properties on the multinomial distribution;

Search for complexes

- Starting from the seeding subgraphs, apply simulated annealing search to expand and identify candidate complexes;

- Output subgraphs with ratio scores exceeding a certain threshold