An Approach to DNA Sequence Classification Through Machine Learning: DNA Sequencing, K Mer Counting, Thresholding, Sequence Analysis

1.1 Background

DNA comprises of two chains of nucleotides spiraled around each other, joined together through hydrogen bonds while moving in diverse directions. It has a double-helix structure, a spiral consisting of two DNA chains coiled around each other (Chou & Shen, 2006). Each of the chains possess four complementary nucleotides – adenine (A), cytosine (C), guanine (G) and thymine (T) (Akhtar et al., 2008),(Akhtar et al., 2008),(Akhtar et al., 2007),(Ramachandran et al., 2012) with an A on one chain always matched with T on the other, and C always matched with G (Kinsner, 2010). The structure of DNA was discovered by Francis Crick. This methodology of expressing gene in the field of biomedical sciences is employed to determine human disease structure (Kirk et al., 2018),(Phongwattana et al., 2015). DNA sequencing is an operation of identifying the state of nucleotides in DNA i.e. nucleic acid sequence. It is the process of identifying the physical order of these bases. There a number of techniques for the identification of the order of four bases. Traditionally most of the biologists use Machine Learning (ML) models to resolve their problems like functional genomics, gene-phenotype associations, gene signatures and gene interactions. Previous research recognizes the genes through experimentation on realistic cells, a veracious but costly job. In contrast the present-day work uses machine based approaches to identify the genes due to inherent accuracy driven advantage of these methods. Machine approaches for gene prediction can be categorized as Content-based approaches and Similarity-based approaches (Wang et al., 2004). Similarity-based formulation search for monotony between candidate and public sequence database of existing genes. Similarity-based formulation are computationally costly and miss original genes. Content-based formulation is advanced technique of gene-prediction that overcomes limitations faced by similarity based technique. These approaches use several attribute of sequences like codon utilization, sequence length and GC content. They then employ supervised learning or applied mathematics approaches to predict whether a read comprises of any genes. ML has been evidentially effective to resolve various problem types like classification, regression and clustering.