Reaction Systems (RSs) were introduced in the field of natural computing as a qualitative model inspired by biological systems. In a RS, each reaction comprises a set of reactants that generate products unless inhibited by reaction inhibitors. We propose a method for analyzing the attractors of a RS model (the states to which the system converges) and for identifying the entities responsible for their attainment. Our approach builds on (i) the construction of a Labeled Transition System (LTS) based on a formal SOS semantics of RSs, and (ii) the application of a slicing method to the trajectories within the LTS. Our model analysis provides new insights with respect to previous studies. We illustrate our methodology on a case study that demonstrates its capability to identify stimulus combinations that lead to specific phenotypes, but also to elucidate the involvement of proteins within T cells in each scenario. These findings allow for a better understanding of the phenomenon and for the identification of potential drug targets for diseases.