Experiment to Find Out Suitable Machine Learning Algorithm for Enzyme Subclass Classification

Proteins play a major role in determining many characteristics and functions of living beings. Prediction of protein classes and subclasses is one of the prominent topics of research in bioinformatics. Machine learning methods are widely used for prediction purposes, also applied for classification and subclassification of proteins. The problem is to classify the proteins to the corresponding subclass they belong to and choose a suitable machine learning method which can be used for better subclass classification. The objective is to compare the performances of three existing machine learning methods: logistic regression, support vector machine (SVM), and random forest, for protein subclassification. For this study the methods are implemented, and their results are compared by varying the number of samples of different subclasses and varying the number of subclasses. Logistic regression and support vector machine are used as a binary classifier for predicting multiple classes with \(log_2(n)\) number of classifiers for n class labels. It is observed that both random forest and support vector machine provide almost same accuracy for smaller data size, but as the data size increases random forest performs better than SVM.