Predição de rotas metabólicas de enzimas utilizando aprendizado de máquina
Carregando...
Data
2018-08-31
Autores
Almeida, Rodrigo de Oliveira
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
Enzimas são uma classe de proteínas responsáveis por catalisar diversos tipos de reações químicas presentes em diferentes rotas metabólicas, sendo assim o principal foco de estudo nas áreas de engenharia metabólica e biologia sintética. Contudo, a anotação de enzimas e a identificação da rota metabólica em que atuam, são frequentemente baseados na similaridade de sequências previamente descritas. A falta e dificuldade de anotação das enzimas se devem pela diversidade funcional em sequências similares de famílias proteicas, sequências espécie-específicas e a dificuldade na definição de homologia em larga escala. De modo a auxiliar a superar tais problemas, o presente trabalho objetivou criar um classificador de rotas metabólicas de enzimas baseado inteiramente nas características da estrutura primária de enzimas e utilizando aprendizado de máquina. A ferramenta computacional criada (mAppLe - Metabolic Pathway Prediction of Enzymes) é composta por 11 preditores de rotas metabólicas de fungos, podendo assim auxiliar nas anotações dos bancos de dados e em trabalhos nas diferentes áreas de pesquisa, como biologia sintética e engenharia metabólica. As performances médias de predição foram de 94% de acurácia, 44% de taxa de falsa descoberta, 67% de F- score , 98% de sensitividade, 93% de especificidade e 0,69 para coeficiente de correlação de Matthews . Com base no desempenho dos preditores criados, constata-se que a ferramenta computacional criada pode ser aplicada com grande sucesso na predição de rotas metabólicas de enzimas de fungos, independente da similaridade das sequências.
Enzymes are a class of proteins that are responsible for catalyzing chemical reactions in numerous metabolic pathways and are often "main targets" in metabolic engineering and synthetic biology. However, enzyme annotation and metabolic pathway identifications are often based on sequence similarities to previously well-described enzymes. Functional diversity in similar sequences of protein families, species-specificity, and difficult-to-define large-scale homologies results in difficulties and a lack of annotation. Here, we present the mAppLe (Metabolic Pathway Prediction of Enzymes), the first metabolic pathway classifier for enzymes based only on primary structure features and a machine learning approach, surpassing limitations imposed by sequence similarities. This tool is composed of 11 pathways predictors for fungi, that can help databank annotations and several type of researches like synthetic biology and metabolic engineering. Results show an average performance of 94% to accuracy, 44% false discovery rate, 67% F-score, 98% sensitivity, 93% specificity and 0.69 to Matthews coefficient correlation. Based on the performance of this predictors, the computational tool created (mAppLe) can be applied successfully to predict pathways of enzymes of the fungi, independent of sequence similarity.
Enzymes are a class of proteins that are responsible for catalyzing chemical reactions in numerous metabolic pathways and are often "main targets" in metabolic engineering and synthetic biology. However, enzyme annotation and metabolic pathway identifications are often based on sequence similarities to previously well-described enzymes. Functional diversity in similar sequences of protein families, species-specificity, and difficult-to-define large-scale homologies results in difficulties and a lack of annotation. Here, we present the mAppLe (Metabolic Pathway Prediction of Enzymes), the first metabolic pathway classifier for enzymes based only on primary structure features and a machine learning approach, surpassing limitations imposed by sequence similarities. This tool is composed of 11 pathways predictors for fungi, that can help databank annotations and several type of researches like synthetic biology and metabolic engineering. Results show an average performance of 94% to accuracy, 44% false discovery rate, 67% F-score, 98% sensitivity, 93% specificity and 0.69 to Matthews coefficient correlation. Based on the performance of this predictors, the computational tool created (mAppLe) can be applied successfully to predict pathways of enzymes of the fungi, independent of sequence similarity.
Descrição
Palavras-chave
Aprendizado de máquina, Bioinformática, Enzimas, Rotas metabólicas, Machine learning, Enzymes, Metabolic pathways