Dynamic Bayesian modeling of the social behavior of Drosophila melanogaster

Abstract

Organismal behavior has always been a challenge to understand. Insects are one of the amenable systems used to understand behavior. A striking variety of insect behaviors gain support from genetic and physiological studies. Drosophila, a widely studied model organism due to its known molecular pathways, has also been popular in behavioral studies. Several behavioral traits in Drosophila including mating, locomotion, and oviposition choice have been traced to the neuronal level. Yet, the results of behavioral analyses are equivocal since they often overlook the external milieu, such as social context, which evidentially influences behavior. There have been many attempts to model Drosophila behavior, however, all have some fundamental issues like lack of complexity, limitation to isolated organisms, and lack of explainability. Here, we model the behavior of a pair of Drosophila melanogaster flies using a novel Dynamic Bayesian Network based approach to better understand behavior in a social context. Two models are proposed, each of which is further used as a predictor for predicting the behavior of both the flies in the pair. They are evaluated on an existing dataset and achieve a remarkable performance: 98.22% and 98.32% accuracy on the two models. Our modeling approach could be applied in predicting animal behaviors in a wide variety of contexts to support existing behavioral studies.