Probabilistic Neural Network for Predicting Resistance to HIV-Protease Inhibitor Nelfinavir

Resumo

Resistance to antiretroviral drugs has been a major obstacle for a long-lasting treatment of HIV infected patients. The development of models to predict drug resistance is already recognized as useful for helping the decision making process regarding the best therapy for each individual HIV+. The aim of this study was to develop classifiers for predicting resistance to HIV protease inhibitor Nelfinavir using probabilistic neural network (PNN). The data were provided by the Molecular Virology Laboratory of the Health Sciences Center, Federal University of Rio de Janeiro (CCS-UFRJ/Brazil). Using a combination of bootstrap and cross-validation to develop the classifiers, four features were selected to be used as input for the network. Additionally, this approach was also used to define the spread parameter of the PNN networks. Final modelling strategy involved the development of four PNN networks with balanced data and evaluation of the models was done using a separate test set. The accu racies on the test set of the classifiers ranged from 71.2 to 76.0% and the area under the receiver operating characteristic (ROC) curve (AUC) ranged from 0.70 to 0.73. For the two best classifiers the sensitivity and specificity were 66.7% and 78.9% respectively, and the accuracy and AUC were 76.0% and 0.73 for both classifiers. The classifiers showed performances very close to two existing expert-based interpretation systems (IS), the Stanford HIV db and the Rega algorithms. The analysis also illustrates the use of a computational approach for feature selection and model parameters estimation that can be used in other settings.