Mineração de dados usando álgebra linear para a predição de alvos drogáveis

Abstract

This work presents the development of a method for recovering target proteins that are druggable. From the representation of drug targets defined as vectors by using InterPro annotations, tools of linear algebra related to singular value decomposition are used to organize the semantic vector space and allow the efficient recovery of proteins related to a given query. Not prima facie relationships arise and indicate drug repositioning opportunities, rational development strategies and, the prediction of potential druggable targets and latent side-effects. The InterPro signatures which are most relevant to drug target/non-drug target discriminating were selected by logistic regression. The results are statistically evaluated by ROC curves analysis and data corroborated in the literature.