The ENCODE projects have generated large high-quality functional genomic datasets which have the potential to dramatically impact our understanding of the specific mechanisms and general principles of the function of cell-specific regulatory elements. We propose to develop an SVM-based computational model to predict enhancers from these datsets and resolve their fine- scale structure. We will utilize an integrative approach to investigate these fine scale features which combines novel computational development, statistical analysis of ENCODE datasets, systematic scoring of human sequence variation, and high throughput validation to improve our understanding of how DNA sequence features and variation contribute to regulatory function. Based on our previous work using k-mer features to predict mammalian enhancers from genomic DNA sequence, we propose improvements in the treatment of sequence features which facilitate statistically robust estimation of long k-mer features and improved spatial resolution. This approach does not rely on previous biological knowledge, and uncovers the sets of novel TFs and cofactors which specify their cell-specific activity. We will train this model on ENCODE DNase-seq and ChIP-seq data and catalogue the regulatory elements in the available human cell-line and mouse datasets. In addition, this model makes specific predictions of the contributions of individual features to enhancer activity, so we propose to experimentally test this set of predictions by directly quantifying the impact of mutation of these elements in a luciferase reporter system. Finally we will evaluate and experimentally assess the predicted impact of specific human SNPs in a set of targeted cell lines. This project should contribute significantly toward a predictive model of regulatory element function and an understanding of how sequence variation impacts disease.
We propose to develop SVM-based models to identify and predict the fine-scale structure of cell-specific regulatory elements from ENCODE datasets, assess the functional impact of human varation on the activity of these elements, and directly evaluate the predictions in experiments in the relevant human cell-lines. This project should significantly contribute to our understanding of regulatory element function and how sequence variation impacts disease.
|Ghandi, Mahmoud; Lee, Dongwon; Mohammad-Noori, Morteza et al. (2014) Enhanced regulatory sequence prediction using gapped k-mer features. PLoS Comput Biol 10:e1003711|