Kmer-Node2Vec: a Fast and Efficient Method for Kmer Embedding from the Kmer Co-occurrence Graph, with Applications to DNA Sequences

Abstract

Current alignment-free DNA sequence comparison tools, such as DNA2Vec and Seq2Vec, have become ubiquitous and are widely used in the computational biology. However, these approaches do not widely apply to the large-scale DNA sequential retrieval due to their extremely long training time. This paper proposes a graph-based DNA embedding algorithm, called KMer-Node2Vec, which converts the large DNA corpus into a k-mer co-occurrence graph, then takes the k-mer sequence samples from this graph by randomly traveling and finally trains the k-mer embedding on this sampling corpus. We posit that because of the stable size of k-mers, the size and density of the k-mer co-occurrence graph change slightly with the increase in the training corpus. Therefore, KMer-Node2Vec has a stable runtime on the large-scale data set, and its performance advantage becomes more and more obvious with the growth of the training corpus. Extensive experiments conducted on real world data sets demonstrate that KMer-Node2Vec outperforms DNA2Vec by 29 times in terms of efficiency in 4G data set, while the error generated by the random walk sampling is small.