Transient complex formation and mutual posttranslational modification of proteins in signaling generate a combinatorial number of reachable chemical species. Traditionally, model reduction heuristics have been applied in kinetic modeling to cope with this complexity. Such unprincipled model reductions are prone to be wrong because they are not based on experimental evidence. Exploiting the local context on which many protein-protein interactions are conditioned, rule-based modeling allows to concisely encode, to simulate and to reduce such combinatorial reaction systems. We present our efforts to transform rule-based models into species reaction systems involving a minimal set of abstract species. Furthermore we propose a framework for model reduction and calibration based on metrics between stochastic processes.