Development of an assessment method for freely moving nonhuman primates' eating behavior using manual and deep learning analysis

Leslie Jaesun Ha; Meelim Kim; Hyeon-Gu Yeo; Inhyeok Baek; Keonwoo Kim; Miwoo Lee; Youngjeon Lee; Hyung Jin Choi

doi:10.1016/j.heliyon.2024.e25561

Development of an assessment method for freely moving nonhuman primates' eating behavior using manual and deep learning analysis

Heliyon. 2024 Feb 5;10(3):e25561. doi: 10.1016/j.heliyon.2024.e25561. eCollection 2024 Feb 15.

Authors

Leslie Jaesun Ha¹, Meelim Kim^{1

2

3

4}, Hyeon-Gu Yeo^{5

6}, Inhyeok Baek¹, Keonwoo Kim^{5

7}, Miwoo Lee¹, Youngjeon Lee^{5

6}, Hyung Jin Choi¹

Affiliations

¹ Department of Biomedical Sciences, Wide River Institute of Immunology, Neuroscience Research Institute, Seoul National University College of Medicine, Republic of Korea.
² Department of Preventive Medicine, Yonsei University College of Medicine, Seoul, Republic of Korea.
³ Center for Wireless and Population Health Systems (CWPHS), University of California, San Diego, La Jolla, CA, 92093, USA.
⁴ Herbert Wertheim School of Public Health and Human Longevity Science, University of California San Diego, San Diego, CA, United States.
⁵ National Primate Research Center, Korea Research Institute of Bioscience and Biotechnology (KRIBB), Republic of Korea.
⁶ KRIBB School of Bioscience, Korea National University of Science and Technology, Republic of Korea.
⁷ School of Life Sciences, BK21 Plus KNU Creative BioResearch Group, Kyungpook National University, Republic of Korea.

Abstract

Purpose: Although eating is imperative for survival, few comprehensive methods have been developed to assess freely moving nonhuman primates' eating behavior. In the current study, we distinguished eating behavior into appetitive and consummatory phases and developed nine indices to study them using manual and deep learning-based (DeepLabCut) techniques.

Method: The indices were utilized to three rhesus macaques by different palatability and hunger levels to validate their utility. To execute the experiment, we designed the eating behavior cage and manufactured the artificial food. The total number of trials was 3, with 1 trial conducted using natural food and 2 trials using artificial food.

Result: As a result, the indices of highest utility for hunger effect were approach frequency and consummatory duration. Appetitive composite score and consummatory duration showed the highest utility for palatability effect. To elucidate the effects of hunger and palatability, we developed 2D visualization plots based on manual indices. These 2D visualization methods could intuitively depict the palatability perception and hunger internal state. Furthermore, the developed deep learning-based analysis proved accurate and comparable with manual analysis. When comparing the time required for analysis, deep learning-based analysis was 24-times faster than manual analysis. Moreover, temporal and spatial dynamics were visualized via manual and deep learning-based analysis. Based on temporal dynamics analysis, the patterns were classified into four categories: early decline, steady decline, mid-peak with early incline, and late decline. Heatmap of spatial dynamics and trajectory-related visualization could elucidate a consumption posture and a higher spatial occupancy of food zone in hunger and with palatable food.

Discussion: Collectively, this study describes a newly developed and validated multi-phase method for assessing freely moving nonhuman primate eating behavior using manual and deep learning-based analyses. These effective tools will prove valuable in food reward (palatability effect) and homeostasis (hunger effect) research.

Keywords: Assessment method; Deep learning-based analysis; Eating behaviors; Hunger; Non-human primate; Palatability.