Complexity of today's systems prevents designers from knowing everything about them and makes engineering them a difficult task for which classical engineering approaches are no longer valid. Such a challenge is especially encountered in actual complex systems simulation in which underlying computational model is very tough to design. A prospective solution is to unburden designers as much as possible by letting this computational model self-build. Adaptive multi-agent systems are the foundation of the fourlayer agent model proposed here for endowing systems with the ability to selftune, self-organize and self-assemble. This agent model has been applied to an application (MicroMega) related to computational biology which aim is to model the functional behavior of unicellular yeast Saccharomyces Cerevisiae.