Behavioral quantification through active learning and multidimensional physiological monitoring

Project Summary/Abstract Naturalistic contexts provide the opportunity to study the brain and behavior in response to the ethological problems an animal is evolutionarily designed to solve. We seek to expand the capabilities of our current behavioral segmentation approaches to provide a more precise and comprehensive account of behavior. By

incorporating recent innovations in machine learning, segmentation approaches that can account for behavioral dynamics at multiple timescales, and increased breadth in the sampling modalities used to classify behaviors, we will create a toolkit that our team and others can make use of to quantify complex, spontaneous behaviors.

We will implement an analysis pipeline to capture and make use of patterns of mouse body position, vocalizations, and arousal states. We also aim to capitalize on recent insights into the role of the gut-brain axis in shaping behavior. After validating our acquisition and analytical approaches, we will monitor these outputs in

response to controlled, parametric environmental manipulations in two distinct, ethologically-relevant contexts: intruder response to resident urine signals and limited access to water. The exploratory data collected in these experiments will be vital to validating our algorithmic advances and for piloting future grant proposals.

The foundation of this work is a diverse team approach. Our team, comprised of experts in social behavior ethology, microbiota research information theory, and data-driven computational modeling, will take an end-to end approach in executing this proposal. By starting with experimental design informed by all parties, we will

ensure that the resulting pipeline possesses sufficient structure and richness for meaningful analysis. the team will help guide long-term research avenues that are both ethologically appropriate and computationally rigorous. Lastly, we recognize that open access will greatly accelerate the validation and adoption of these technologies,

a stated aim of this RFA. Dissemination and access to our deliverables will benefit substantially from ongoing relationships with the Pittsburgh Supercomputing Center and OpenBehavior. The partnered hardware and software advances of Aim 1a and 1b represent the overarching goal of this proposal, an advanced and comprehensive behavior segmentation platform. Aim 2 will interrogate temporally-

dynamic urine protein signals and Aim 3 will study how progressively increasing thirst induced through water- restriction affect neurobehavioral measures. These contexts will be used to benchmark the broad applicability of Aim 1 – as well as to explore the potential to address targeted research questions within these frameworks.