Optimization of Spaced K-mer Frequency Feature Extraction using Genetic Algorithms for Metagenome Fragment Classification

Abstract

K -mer frequencies are commonly used in extracting features from metagenome fragments. In spite of this, researchers have found that their use is still inefficient. In this research, a genetic algorithm was employed to find optimally spaced k -mers. These were obtained by generating the possible combinations of match positions and don’t care positions (written as *). This approach was adopted from the concept of spaced seeds in PatternHunter. The use of spaced k -mers could reduce the size of the k -mer frequency feature’s dimension. To measure the accuracy of the proposed method we used the naive Bayesian classifier (NBC). The result showed that the chromosome 111111110001, representing spaced k -mer model [111 1111 10001], was the best chromosome, with a higher fitness (85.42) than that of the k -mer frequency feature. Moreover, the proposed approach also reduced the feature extraction time.