This article presents a new formalism Interac-DEC-MDP whose aim is to introduce the concept of interaction in Decentralized Markov Decision Process and which has been inspired by biology. The aim of this formalism, Interac-DEC-MDP, is to describe and represent interactions among agents. The outcome of interactions is decided collectively by two agents and is in charge of the distribution of local rewards. We have modeled a biological experiment within this formalism. A simple learning algorithm applied on this formalism generates a more efficient collective behavior than without interactions.