Enhancer prediction using distance aware kernels

The regulation of gene expression is important for the development of living cells and their responses to environmental conditions. This mechanism is controlled, to a large extend, by transcription factors that bind to regulatory sequences, such as enhancers. The identification of enhancers is therefore important for understanding the regulatory networks within cells. In this paper, we propose new features and kernels that can be used with support vector machine (SVM) classifiers to predict enhancers from genomic sequences. These are based on general sequence features and kernels but are extended to incorporate the information about the distance between the features, thus can better capture the spatial preferences and combinatorial binding rules of transcription factors. Experiments on predicting enhancers in human and Caenorhabditis elegans show that, by combining the proposed features and kernels with SVM, our method achieves state-of-the-art accuracy and outperforms a leading enhancer prediction method.