Ethoflow: an artificial intelligence-based software that facilitates behavioral measurements and their application for toxicological assessments in insects

• Ano de defesa: 2021
• Autor(a) principal: Bernardes, Rodrigo Cupertino
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Viçosa
• Palavras-chave em Português: Abelhas - Comportamento; Ethoflow (Software); Entomologia Agrícola
• Link de acesso: https://locus.ufv.br//handle/123456789/28974

The application of artificial intelligence (AI) techniques has demonstrated outstanding performance for automating complex tasks in many areas. Thus, software with AI implementation can be sufficiently robust to meet the demand for studies on animal behavior (e.g., evaluating behavior under field conditions). Studies on risk assessment in bees have focused on the possible causes of loss of their colonies worldwide, which is attributed to different factors, including agricultural practices with the use of agrochemicals. In toxicological studies of agrochemicals in bees, behavioral assessment is an important sublethal parameter. Thus, the development and application of AI tools can considerably contribute to the understanding of how agrochemicals and other factors are affecting the bees’ health. The present work aimed to develop an AI-based software (Ethoflow) to automatically assess animal behavior. In addition, Ethoflow was applied to evaluate sublethal behavioral changes in toxicological studies of agrochemicals in forages of two stingless bee species, including Melipona quadrifasciata and Partamona helleri (Hymenoptera, Apidae, Meliponini). The results obtained demonstrate that: (1) Ethoflow is robust for multivariate behavioral assessments, behavioral assessments in heterogeneous environments, tracking individuals in groups maintaining their identities and can be trained to learn behaviors specific to animals; (2) it is possible to classify agrochemical contamination in bees with high accuracy by integrating multivariate behavioral data with AI algorithms and the agrochemicals glyphosate and imidacloprid differentially impact the midgut physiology of M. quadrifasciata; (3) the foliar fertilizer copper sulfate (CuSO 4 ) causes sublethal effects on the behavior and structure and physiology of the midgut epithelium of P. helleri; (4) the mixture of mesotrione and atrazine herbicides interfered in food intake and behavioral parameters, caused damage to the midgut epithelium and altered the pattern of proteins related to the cell proliferation and differentiation in midgut of P. helleri. In general, Ethoflow is a useful support tool for technical-scientific applications in the animal behavior field and has significant potential in risk assessments of non-target organisms for modeling the multiple factors affecting bees’ health, including theadverse effects of agrochemicals. Besides, AI algorithms trained with multivariate behavioral data predict bees’ agrochemical contamination with high accuracy. Finally, the analyzes enabled holistic assessments of sublethal effects of different agrochemicals on the behavior and physiology of bees. Keywords: Bee. Computer vision. Machine learning. Meliponini. Pollinators.