Gene Sequence Classification Using K-mer Decomposition and Soft-Computing-Based Approach

Abstract

The healthcare industry is moving toward personalized medicine which requires the use of individual genetic information so that medical treatment can be customized to the specific properties of an individual. DNA sequence of a genome consists of several genes. These genes are the basic building blocks of an organism. A human genome consists of 20–30 thousand genes. Some of these genes are involved in the growth and development of the body and some are responsible for the production of critical diseases (influenza, ebola, dengue) remaining are the non-coding (junk) genes. Identification and classification of these genes into a few biological meaningful groups: coding, non-coding, and viral are useful for the treatment and diagnosis of an organism. In this paper, k-mer (substrings of length k) frequency decomposition and soft-computing-based approach is used to classify and identify the large set of unknown genes into some meaningful groups. It works by first taking a DNA sequence and computing a vector of the proportions of every possible k-mer. These vectors are used as feature vectors, and a well-known supervised classification algorithm (multi-mode Naive Bayes classifier) is trained on the vectors. Experiments show that the proposed approach achieves the highest accuracy ($\ge$90 $\%$) along with the lowest running time in comparison with other state-of-the-art methods.