Prediction of alpha helices in proteins using Modified Logistic Regression Model

• Autor(a) principal: Carmelina Figueiredo Vieira Leite
• Data de Publicação: 2016
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/33892
• topic: Logistic regression; Prediction; Protein; Structure; Bioinformática; Modelos Logísticos; Previsões; Proteínas

The advance in proteins secondary structure prediction produces directly impacts on health and biological processes knowledge. Despite the achievements and advances, the prediction of proteins structure remains a challenge. Considering this fact, we propose a de novo method for the prediction of alpha helix. Initially, we created a list of proteins with low identity between them, from the repository Protein Data Bank, using PISCES. Each protein was separated into fragments (of size 9) using the sliding window technique. From the obtained fragments, we classified them into the ones that were 100% a standard type alpha helix, the ones that were not a 100% of the same type of secondary structure. For each fragment, we used a sliding window of size 3 to characterize them. These had a value associated with the occurrence of the alpha helix structure. It was possible to predict the secondary structure group, alpha helix, of an unknown protein/query. To accomplish our goals, we used modified logistic regression and constructed two methods for prediction of these structures. Tests of accuracy and specificity applied to the methods gave results greater than 70%. Unfortunately, the sensitivity did not show good results. One of the methods revealed to be a very promising application for the secondary structure prediction problem, and to a possible usage in other purpose. All methods were implemented in MatLab R2015b (2015)