An Approach Based Natural Language Processing for DNA Sequences Encoding Using the Global Vectors for Word Representation

Abstract

DNA sequence has several representations; one of them is to split it into k-mers components. In this work, we explore the high similarity between natural language and “genomic sequence language” which are both character-based languages, to represent DNA sequences. In this representation, we processed a DNA sequence as a set of overlapping word embeddings using the Global Vectors representation. In Natural language processing context, we can consider k-mers as words. The embedding representation of k-mers helped to overcome the curse of dimensionality, which is one of the main issues of traditional methods that encode k-mers occurrence as one hot vector. Experiments on the first Critical Assessment of Metagenome Interpretation (CAMI) dataset demonstrated that our method is an efficient way to cluster metagenomics reads and predict their taxonomy. This method could be used as first step for metagenomics downstream analysis.