Effective DNA Encoding for Splice Site Prediction Using SVM

Abstract

Splice site prediction in the pre-mRNA is a very important task for understanding gene structure and its function. To predict splice sites, SVM (support vector machine)-based classification technique is frequently used because of its classification accuracy. High performance of SVM largely depends on DNA encoding method. However, existing encoding approaches do not reveal the characteristics of DNA sequences very well enough to provide as much information as sequences have. In this paper, we propose new effective DNA encoding method for feature extraction which can give more information of DNA sequence. Our encoding method can provide density information of each nucleotide along with positional information and chemical property. Extensive performance study shows that the proposed method can provide better performance than existing encoding methods based on several performance criteria such as classification accuracy, sensitivity, specificity and auROC (area under receiver operating characteristicscurve).