Discovering distal regulatory elements by integrating multiple types of chromatin state maps

While the annotation of human protein-coding sequences is now fairly comprehensive, the identification of regulatory sequences remains difficult. With higher resolution, fewer artifacts and greater coverage, short-read-sequencing-based technologies have made striking impact on genome research. Given the rapid accumulation of genome-wide chromatin state data, there is a pressing need for computational methods to analyze these data. In this paper, we developed a Multiple Layer Perceptron (MLP) framework to predict transcriptional enhancers by integrating multiple types of chromatin state maps, including histone modifications, DNase I cleavage, and DNA methylation. Comparisons with previous work using known enhancers from three cell types suggest that our algorithm is more robust and has higher precision and sensitivity. We anticipate that the new method will be a valuable tool for genome-wide mapping of various DNA regulatory elements in a wide variety of cell types, tissues and growth conditions.