Development and performance appraisal of an expert system for predicting HCV genotype using a virtual restriction map and artificial neural network

Background: Hepatitis C virus (HCV) is a positive strand RNA virus belonging to the Flaviridae family of viruses with an approximately 9.4 kb genome size, which was discovered in 1989. HCV is classified into types, corresponding to the main branches in the phylogenetic tree; and subtypes, corresponding to the more related sequences within the major groups. Our objective was to accurately predict various HCV types and subtypes from a complex set of restriction site patterns by the use of artificial intelligence system. Three methods for classifying of genes based on artificial neural network (ANN), Principal component analysis (PCA) and support vector machine (SVM) were proposed. Material and methods: These methods classify genes into groups which are made distinct from each other by evolutionary changes in their restriction sites. Classification models were constructed based on changes at restriction site in the untranslated region (UTR) of the virus. Models were trained, validated, and tested with 330 UTR sequences. A procedure of determining the optimal network parameters was proposed to speed up the training process. Result and Discussion: The results suggested that the restriction site map was a more accurate predictor of HCV genotype. To show the ability of the model in genotyping, the internal representations developed by the networks were analyzed by principal component analysis. This analysis showed that the networks are able to discover relevant features just on the basis of the association between the restriction map and the virus genotype.