We describe a heterogeneous neural network where neurons interact by means of various neurotransmitters using the common extracellular space. Every neuron is sensitive to a subset of neurotransmitters and, when excited, secretes its specific neurotransmitter. This feature enables establishing the selective connections between neurons according to sets of their receptors and to their outputs. We use a simplification of this formalism as a basis for modeling interactions between agents in a social network, where the two opposite types of activity are spreading. Agents have beliefs of different strength and activation thresholds of different heights (which correspond to neuronal excitation thresholds) and can be more or less sensitive to an external influence (which corresponds to weights of neuron receptors). The main properties of the agents and the principles of activity spreading are defined. The classification of agents according to their parameters is provided.