Lee, S.; Fu, Z.; Choi, S.

Traditional open-field assays quantify rodent behavior using low-dimensional scalar metrics that often overlook dynamical and temporally dependent aspects of spontaneous behavior. Here, we establish a high-throughput framework that integrates multi-camera 3D pose estimation from AVATAR-3D (1) with unsupervised autoregressive hidden Markov modeling using Keypoint-MoSeq (2-4) to transform recordings into time sequences of discrete, recurring behavioral units termed syllables. Drawing inspiration from information theory (5) and spectral analysis of Markov systems (6,7), we compute Shannon entropy and the second-largest eigenvalue (Eigen2) of the syllable transition matrix. Entropy and Eigen2 offer complementary lenses through which spontaneous behavior can be understood - one capturing dispersion of behavioral expression, the other reflecting temporal persistence and mixing dynamics. Together, these metrics quantify the organization of behavior as a stochastic dynamical system rather than a collection of isolated actions.