Automatic Classification of Genetic Mutations Based on Machine Learning Methods

УДК 519.687:004.912

Abstract

This paper considers the problem of identifying the type of genetic mutation of a cancer tumor after sequencing its genome. The problem solution relates to multi-class classification problems. The paper proposes an approach for the identification of mutation classes based on their text description using supervised machine learning methods. The study is carried out using a data set on cancer diseases obtained by the analysis of genome mutations in tumor cells. The data set includes the gene type, its mutation, a text description of the gene mutation, and the mutation class (with nine different classes overall). The paper provides an analysis and justification of the optimal text preprocessing and vectorization methods suitable for the source data at the inputs of the machine learning methods. There are several text data classifiers based on the k-nearest neighbors, decision trees, Bayesian classifiers, and the logistic regression machine learning methods. The obtained classification performance metrics after many simulations reveal that the best method to perform classification and identification of mutation classes is the linear regression method with the lowest error rate.