(a) 3D Genome. The 3D shape of chromatin can lead DNA ‘windows’ (shown in gray boxes) far apart in the 1D genome space to be spatially close. These spatial interactions can influence chromatin profiles, such as TFs binding (as shown by the colored shapes). In most cases, the DNA sequence determines the chromatin profile. However, it can also be influenced by interactions, such as the formation of TF complexes shown in the middle. (b) Graph Representation of DNA. Using Hi-C data, we can represent subfigure (a) using a graph, where the lines between windows are the edges indicated by Hi-C data. (c) ChromeGCN. By using a GCN on top of convolutional outputs the model considers the known dependencies between long-range DNA windows. The lines between windows correspond to edges in Hi-C data