DNA classifications with self-organizing maps (SOMs)

The objective of this paper is to apply self-organizing maps (SOMs) for exon/intron classification in DNA using windowed splice junction data. Splice junctions are group of nucleotides that serve as boundaries between sections of DNA that code for genetic material and sections that do not. Genes are often interrupted by sections of non-coding DNA sequences. The data used for this study is human DNA data taken from the National Center for Bioinformatics Information (http://www.ncbi.nih.gov/). The DNA dataset contains 1,424 DNA sequences with 128 descriptors for each sequence. SOMs are used to classify each DNA sequence into three categories that are sequences of transition from gene (exon) to non-gene (intron), non-gene (intron) to gene (exon), and no transition categories where the two-base pair code for the splice junction was coincidental. The multidimensional sequences are clustered into a two-dimensional space that was graphically displayed for data exploration and classification. Visual and graphical capabilities of SOMs are applied to classify the DNA dataset. The topographic properties of SOMs preserve similar sequences close to each other on the output map. Clusters of the dataset are determined and labeled based on the classes of the output neuron in the cluster. The highest frequency classes mapped on the output neuron are labeled as the classes of the output neuron.