Exploring Equivalence Metrics to Analyze Behavior in a Naked Mole Rat Colony

Many statistical and data mining techniques have been used to analyze the deluge of data generated by computerized, sensing devices. Behavioral psychologists traditionally have relied on "low-tech" methodologies for observing animal behavior in the wild and the laboratory. These methods are time intensive and laborious. When the observed animal is a colony animal, with many individuals to observe, traditional methods fail We inject RFID passive transponders under the skin of our study animal, the Naked Mole-rat (NMR). RFID readers are placed throughout the housing environment, allowing us to track animal movements as they move through these areas, with sub-second resolution for long periods of time. This methodology generates huge amounts of data requiring Big Data analytical techniques. In this paper, we investigate equivalence metrics, specifically the Pearson Correlation Coefficient and Hamming Distance, to analyze behavior changes in the social network structure of a naked mole rat colony. Our results showed that a Pearson Correlation was sufficient to detect equipment error and Hamming Distance could detect changes in colony behavior. Background and Motivation Traditional tools used to measure behavior of laboratory animals become problematic when large groups of animals need to be measured simultaneously. Our recent work has examined social behavior using network tools in a highly specialized laboratory-housed rodent, the African naked mole-rat (Heterocephalus Glaber). African mole-rats are unique among rodents, indeed among all mammals, because they participate in a cooperative breeding eusocial lifestyle. Naked mole-rats rely on a queen to be solely responsible for breeding the entire colony. The lack of reproductive ability, and close kin relationships among colony members, allows the naked mole-rat to maintain the largest colony size of any mammal with up to 300 memCopyright © 2015, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. bers in wild colonies, and up to 100 animals in captive colonies. Naked mole-rats are quickly becoming an important model in biomedical research. The naked mole-rat genome has more genes in common to humans than traditional laboratory rodents (rats and mice; Kim, et al. 2011), yet they demonstrate cancer and stroke resistance (Larson & Park, 2009) and extreme longevity, living over 30 years (Buffenstein, et al. 2008). These animals thrive in captivity and allow for the study of their cooperative and social behavior. Understanding the brain mechanisms driving the social behavior in the naked mole-rat may help us to further understand human social behavior and social behavior disorders such as autism and schizophrenia. We have presented the first colony-wide analysis of activity patterns and social behaviors of naked mole-rats using customized radio frequency identification (RFID) based tracking system (McCloskey et al. 2011). Using this system, we can track all movements of all animals as they move through all areas of their laboratory habitat with subsecond resolution for long periods of time. The eventbased dataset generated by this type of system allows for the simultaneous study of individual, sub group, and whole group characteristics. Our initial studies in this area were focused on identifying social behaviors of animals in a colony based on the premise that animal interaction can be determined by measuring how much time is spent in the same area of the housing environment. We found that using a combination of data mining approaches, including adjacency matrix sampling, principal component analysis, and frequent pattern mining, we can successfully identify characteristics of the social network (McCloskey et al. 2011). We have also used the RFID data to demonstrate that naked mole-rats, like all animals, have a periodicity to their behavior. The circadian cycle of activity is roughly 24 hours in the presence of external cues such as light and interaction with experimenters and animal care staff. 422 Proceedings of the Twenty-Eighth International Florida Artificial Intelligence Research Society Conference