Crossing Borders: Modeling Approaches to Understand Chromatin Domains and Their Boundaries

Developments in chromosome conformation capture and sequencing technologies have led to an ongoing revolution in the quantities and varieties of data available to study nuclear architecture. However, this new avalanche of data creates new challenges, particularly in generating a meaningful, integrated view of these heterogeneous datasets, and understanding their biological significance. Here we show how modeling approaches borrowed from the field of machine learning can accurately recapitulate known higher order chromatin structure, using only data representing features (histone modifications, DNA binding proteins, etc.) measured at a much lower level. We provide practical information on the construction of such models, from raw data processing to the measure of accuracy and interpretation. The component features "learned" by successful models can provide functional insights into the systems underlying chromosome structure. The integrative view of higher and lower level structural features provided also provides a basis for an exploration of regulatory domain structure, and the boundary elements that define them. We discuss strategies for robust compartment domain calling and metrics for the significance of boundary feature enrichments, and promising future avenues for further research.