Estimação de atividade e propriedades físico-químicas de compostos antibacterianos utilizando aprendizado profundo
Ano de defesa: | 2023 |
---|---|
Autor(a) principal: | |
Orientador(a): | |
Banca de defesa: | |
Tipo de documento: | Dissertação |
Tipo de acesso: | Acesso aberto |
Idioma: | por |
Instituição de defesa: |
Universidade Federal de Minas Gerais
Brasil ENG - DEPARTAMENTO DE ENGENHARIA ELÉTRICA Programa de Pós-Graduação em Engenharia Elétrica UFMG |
Programa de Pós-Graduação: |
Não Informado pela instituição
|
Departamento: |
Não Informado pela instituição
|
País: |
Não Informado pela instituição
|
Palavras-chave em Português: | |
Link de acesso: | http://hdl.handle.net/1843/60719 |
Resumo: | One of the essential elements for maintaining people's health and well-being is the research and development of new drugs. Pursuing innovative compounds and repositioning existing compounds enable the treatment of diseases and improve the quality of life. However, developing new drugs is time-consuming (up to 10 years) and expensive (costing up to 3 billion dollars). In this context, a class of drugs that requires an urgent demand for new developments are antibacterials due to bacteria's significant growth of resistance to antibiotics. Resistant bacterial infections cause higher medical costs, prolonged hospital stays, and increased mortality. Computational tools include approaches that, in addition to accelerating and reducing process costs, mitigate the spread of diseases, including infections caused by resistant bacteria. This computational assistance is used to automate tests and reduce the number of compounds needed in preclinical tests and the initial clinical phases (phases with higher discontinuation rates), focusing resources on the most promising samples. Thus, this study aims to propose models that employ deep learning to predict antibacterial activities and various physicochemical parameters of substances to uncover potential antibacterial agents. Biological activity data of drugs against four Gram-negative bacteria (Escherichia coli, Acinetobacter baumannii, Pseudomonas aeruginosa, and Salmonella typhimurium) were collected, along with three physicochemical properties (water solubility, DMSO solubility, and lipophilicity). Graph Neural Network architecture was employed to address classification and regression tasks. The models underwent a process of hyperparameter optimization. Among other validation metrics evaluated for the models, the results demonstrated an accuracy exceeding 0.70 for classification models and a coefficient of determination above 0.80 for regression models. Compounds with antibacterial activity and more promising physicochemical properties may be experimentally evaluated as potential antibacterials. |