Modelos de aprendizado profundo para avaliação de toxicidade aguda de compostos químicos em aves
Nenhuma Miniatura disponível
Data
2022-08-24
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Goiás
Resumo
The modernization of agriculture has provided economic growth as a result of increased productivity. However, this sector has the intense use of pesticides as a practice, which presents a potential risk to the environment and organisms that provide ecological services such as avian species. Considering that birds are seed-dispersing organisms, in vivo acute toxicity studies have been widely used as regulatory criteria for the registration of new pesticides. However, these studies are usually time consuming, expensive and involve ethical issues. Therefore, this work aimed to develop Quantitative Structure Activity Relationship (QSAR) models, based on machine learning, to predict the acute toxicity of chemical compounds in several bird species. Initially, the compilation, integration and preparation of the largest datasets of compounds with data on experimental toxicological properties were performed for the following avian species: A. platyrhynchos, C.virginianus, C. japonica and P. colchicus. Then, a chemical space analysis showed that the prepared datasets share chemical information with each other, and the correlation of toxicological data between species proved to be moderate (with 'r' around 0.68). At the end of this process, QSAR models for regression tasks were generated using Deep Learning methods. Among them, a multitask model based on Feed-Forward Neural Networks (FFNN), capable of predicting the acute toxicity (pDL50) of several bird species simultaneously, was the most predictive, obtaining r values between 0.59 – 0.80 for the test set. The results demonstrate that the multitasking model was able to promote inductive transfer of learning between tasks, that is, between bioassays of each species. Therefore, the generated model represents a new alternative method to the use of animals for the evaluation of acute avian toxicity.
Descrição
Palavras-chave
Ecotoxicologia, Multitask, QSAR, Ecotoxicology, Multitask, QSAR
Citação
RAMOS, Gabrielle Santos. Modelos de aprendizado profundo para avaliação de toxicidade aguda de compostos químicos em aves. 2022.33 f. Trabalho de Conclusão de Curso (Bacharelado em Farmácia) - Faculdade de Farmácia, Universidade Federal de Goiás, Goiânia, 2022.