A Computational Approach for Detailed Quantification of Mouse Parenting Behavior

Abstract

A major aim of neuroscience is understanding the brain circuitry underlying observable behaviors. Recently, an increasing number of tools have allowed researchers to investigate brain circuitry in unprecedented ways. However, behavior quantification methods have remained coarse and labor intensive, thereby limiting our understanding of underlying brain circuitry. Naturalistic behaviors like parenting are especially difficult to quantify, as they must often be reduced in a laboratory setting and usually lack trial structures. To address this problem, we have created a three-pronged approach to quantify previously described as well as newly defined behaviors and motor actions of parenting in mice. Our first approach is a user-defined system of geometric feature descriptions that were used to quantify 12 previously described parenting behaviors such as pup retrieval, pup investigation, etc. Our second and third approaches uses unsupervised methods like Generalized Linear Model-Hidden Markov Models (GLM-HMM) and Hidden Markov Models (HMM) to quantify newly described parenting behaviors. Firstly, we used a GLM-HMM to identify 3 hidden states underlying the interaction of the adult and pup mouse. Next, we used an unsupervised clustering approach combined with an HMM to identify 14 behavioral states such as forward locomotion, stretch and contract, idle with no head movement, etc. To demonstrate the utility of this behavior quantification framework, we used it to distinguish parenting behaviors in mothers, fathers, and virgin female mice. We also used it to identify behaviors encoded by parenting-related neurons in the medial preoptic area (MPOA). This framework ultimately provides a robust method to quantify naturalistic behaviors.