Title: Comparing behavioral dynamics between groups using hierarchical hidden semi Markov models
Authors: Emmeke Aarts - Utrecht University (Netherlands) [presenting]
Abstract: Technological advances make it increasingly easy to collect intensive longitudinal data on multiple subjects or animals. Hidden Markov models (HMM) are becoming an increasingly popular method to summarize these behavioral data over time. However, when using conventional HMMs on behavioral data, there are two drawbacks. First, HMMs are not well suited to simultaneously model sequences of data of multiple subjects, or to formally compare parameters between groups of subjects. Second, a HMM assumes that the amount of time spent within a hidden state is a function of a memoryless process. However, when investigating behavior over time, this is biologically implausible. We develop and implement a hierarchical hidden semi Markov model (HHSMM) to describe - and formally compare - the temporal organization of behavior over groups. In our model, the state durations are explicitly modeled (i.e., an explicit duration HMM), and a Bayesian framework is used for parameter estimation. We illustrate our proposed model using a real data example, comparing the behavioral pattern of young adult and aged C57BL/6J mice. Our proposed framework is one of the first that models the behavior of multiple animals simultaneously, taking a HSMM approach while allowing for heterogeneity in - and formal group comparisons on - all model parameters.