Implementasi Algoritme Support Vector Machine Dan Fitur Seleksi MRMR Untuk Prediksi Glikosilasi

Abstract

During the protein formation process, there are post-translational modifications that provide additional properties and produce new R groups in the polypeptide chain. One of the results of post-translational modifications is glycosylation. Glycosylation reactions occur between protein and glucose at high concentrations. There are 3 categories of glycosylation found in the human body, namely N-glycosylation, O-glycosylation and C-glycosylation. To determine the functional role of glycosylation, that is by predicting the substrate of each glycosylation site. A computational approach is a way to predict the glycosylation site, using the Support Vector Machine (SVM) algorithm. In this study there are 2 types of data, namely independent data and benchmark data. The features used are feature extraction and feature selection using Maximum Redundancy Minimum Relevance (MRMR) of 25, 50 and 75 columns. SVM classification test using 5-fold cross validation. The highest accuracy result lies in the use of the 75 column MRMR selection feature. In Independent N data, the greatest accuracy lies in the sigmoid kernel with a causation value of 86.66%, while for independent C data, the accuracy is 87.5% in the sigmoid kernel and for independent O data, the largest accuracy is 89.31% which is in the RBF kernel. For benchmark N data, the highest accuracy is 70.54% in the RBF kernel, for benchmark C data the greatest accuracy lies in the RBF kernel with a value of 95.06% and for benchmak O data it is in the RBF kernel with the greatest accuracy, which is 92.64%.