Machine learning approach for classifying histone modifications

Post-translational modifications pegged on to the N-terminal tails of the nucleosomes core histone proteins determine the transcriptional activity of that chromosomal region leading to the histone code hypothesis. We rely on recently produced experimental data on genome-wide maps of chromatin state to derive computational models delineating the hidden patterns of post-translational modifications. We use support vector machines to model and predict post-translational modification state of the histones using a diverse set of features. The classification results on the human CD-4+ T-cells by five-fold cross-validation show promising results. Our analysis reveals better classification accuracy when employing chromosome-specific classifiers. Finally, we show how computational models can be used to provide an initial estimate of the chromatin state and its effect on DNA loci.