To analyse large models using asynchronous updates, one can alternatively randomly pick a single component to be updated at each step, leading to a single successor. This "random walk" in the asynchronous STG is repeated to create a sampling of the reachable attractors. More generally, such random walks can be applied to any updating policy which yields multiple successors.
Availability
BoolNetR package for the simulation, reconstruction and analysis of various types of Boolean networksMaBoSS
Continuous time Boolean modeling