Comparative analysis of performance K-nearest neighbor and weighted K-nearest neighbor combined with global encoding in predicting protein-protein interaction human immunodeficiency virus type 1 with human

Abstract

Human Immunodeficiency Virus type 1 (HIV-1) is the most common type of virus that could cause Acquired Immunodeficiency Syndrome (AIDS). This virus attacks by interacting between HIV-1 proteins and human proteins. This study used data in the form of amino acid sequences of proteins to be selected and then changed its features using global encoding. Then the feature extraction results will be used as input for the k-nearest neighbor models and weighted k-nearest neighbor models to predict the interaction of HIV proteins with human proteins. In addition, this study also compared the evaluation of the k-nearest neighbor (KNN) and weighted k-nearest neighbor (WKNN) models. Then both models are given Manhattan distance and Euclidean distance. The results of this study indicate that the best model is WKNN model with Manhattan distance using parameter global encoding (L = 2) and parameter model (K = 2), obtained the highest evaluation results in predicting the interaction of HIV proteins with human proteins, with accuracy reaching 93.78 %, sensitivity reaching 93.58 %, specificity reaching 94.01 % and precision reaching 95.78 %. Also in this study 193 proteins from 6,046 human proteins were found interacted with HIV-1 and 5853 human proteins from predictive datasets that did not interact with HIV-1.