Prediction of histone modifications in DNA sequences

Abstract

DNA molecules are wrapped around histone octamers to form nucleosome structures whose occupancy and histone modification states profoundly influence the gene expression. Depending on the DNA segment that a nuleosome incorporated, its histone proteins exihibit paticular modifications by added some functional chemical groups to specific amino acids. The key approach up to now to determining the DNA locations of histone occupancy as well as histone modifications is an experimental technique called ChiP-Chip, or Chromatin Immunoprecipitation on Microarray Chip. This experimental technique has some disadvantages such as it is tedious, wastes time and money, produces noise, and cannot provide results at an arbitrarily high resolution, especially with large genomes like human's. We have developed a computational method to determine qualitatively histone-occupied as well as acetylation and methylation locations in DNA sequences. The method is based on support vector machines (SVMs) to learn models from training data sets that discriminate between areas with high and low levels of histone occupancy, acetylation or methylation. Our computational method can give quickly the prediction at any position in a DNA sequence based on the content and context of the subsequence around that position. The prediction results on the yeast genome by three-fold cross-validation showed high accuracy and were consistent with the ones from experimental methods. Moreover, SVM-classification models in our method can present genetic preferences of DNA areas that have high modification levels. ©2007 IEEE.