Graph ensemble null models. The graph null models are described in the four figures shown here. Each graph is generated from a fixed empirical graph (CORE, DIP or LC) according to one of the four different algorithms. (a) and (b) permute purely network structural information while (c) and (d) constrain the generation of random graphs also according to biological constraints (such as GO ontology information and complex annotations). (a) Node shuffle: The labels for each node are permuted (e.g. node colour) but the topology of the graph is fixed. (b) Network shuffle: The degree of each node, [A, B, C, D, E], is fixed along with the node characteristic, colour, whilst the edges are randomly rewired. (c) Bipartite Node Shuffle: This permutes each node to another node, v
such that β(v
) = β(v
) for the particular characteristic, β, under consideration. (d) Bipartite Network Shuffle: This retains the degree of each node, d
(v), and also rewires each edge, e, to one of the available node pairs that share the same edge characteristic, ϕ(e).