Noise is ubiquitously present in cellular networks. In the deterministic limit, noise
 can be treated as a small perturbation but as the particle number decreases discrete noise
 generates complex behavior that cannot be taken into account by the deterministic equation.
 New phenomena such as non-Gaussion distribution and resonant signaling are observed
 and explained. For noise propagating in an enzymatic cascade, we designed a novel
 variational technique to compute the time-dependent distribution. The method can be generalized
 to coping with arbitrary biological networks.