Sci Rep. 2026 Jun 3. doi: 10.1038/s41598-026-55331-0. Online ahead of print.
ABSTRACT
Animal social behavior, such as agonistic behavior, has been extensively studied for a long time since it is one of the key factors that influence the survivability and reproductive success in many animals, including crickets. Gryllus bimaculatus, a field cricket, and Acheta domesticus, a house cricket, are two cricket species whose social behaviors have been extensively studied. However, the interspecific behaviors of these crickets had not been assessed and comprehensively compared to their behaviors when encountering their conspecifics. Moreover, most of the behavior studies in crickets still rely on traditional observational methods, which are labor-intensive and prone to observer bias. Fortunately, due to the advancement in machine learning, a markerless pose estimation method has revolutionized behavioral analysis in animal studies by enabling precise, automated tracking of body parts without the need for physical markers, reducing stress and allowing for naturalistic behavior in animals. Therefore, the study aimed to demonstrate a comprehensive methodology in analyzing the intra- and interspecies social behaviors of G. bimaculatus and A. domesticus by DeepLabCut (DLC), a deep learning-based pose estimation tool. Based on the results, the trained model that was used in the current study showed a relatively high accuracy after undergoing several rounds of training, as indicated by the relatively high value of average likelihood (0.94) and low value of average low likelihood percentage (3.85%) of each detected body part in every tested cricket. Meanwhile, GLMM analysis revealed no significant differences between the proposed method and manual scoring in both calculated aggressive-related behavior endpoints. Moreover, the system achieved high sensitivity with an average of 0.8, with acceptable precision with an average of 0.7 for both behaviors, demonstrating that the automated approach provides accurate and reliable quantification of aggressive interactions in paired crickets. Next, in terms of the cricket’s social behaviors, males of G. bimaculatus displayed aggressive behaviors as indicated by statistically high overt physical combat count, which were followed by high locomotion and movement complexity, while a slightly higher posterior-oriented interaction count was observed in the male × female group of this species. Meanwhile, although overt physical combat behaviors still could be observed in males of A. domesticus, this cricket in other gender combinations did not display noticeable aggressive behaviors compared to G. bimaculatus. On the other hand, aggressive behaviors were still shown by both males and females during interspecific tests, though in a relatively lesser magnitude compared to intraspecific G. bimaculatus. Taken together, the present study highlighted the effectiveness of DLC for markerless pose estimation of crickets, emphasizing the potential of this methodology as an alternative to provide novel quantitative insights into the dynamics of their social interactions, and thus, paving the way for large-scale, reproducible research on animal social interactions that could advance our understanding of animal behaviors.
PMID:42236810 | DOI:10.1038/s41598-026-55331-0
