Prediction of insertion-site preferences of transposons using support vector machines and artificial neural networks

Abstract

Transposons are segments of DNA that are capable of moving from one location to another within the genome of a cell. Understanding transposon insertion-site preferences is critically important in functional genomics and gene therapy studies. It has been found that the deformability property of the local DNA structure of the integration sites, called Vstep, is of significant importance in the target-site selection process. We considered the Vstep profiles of insertion sites and developed predictors based on Artificial Neural Networks (ANN) and Support Vector Machines (SVM), and trained them with a Sleeping Beauty transposon dataset. We found that both ANN and SVM predictors are excellent in finding the most preferred regions. However, the SVM predictor outperforms the ANN predictor in recognizing preferred sites, in general.