Fig. 2

A deep learning model for predicting whether a region of open chromatin exhibits IR. The model receives as input the sequences of intragenic DHSs labeled as associated with IR or non-IR; the one hot encoding is processed through three layers of convolution, followed by three fully connected layers and the output layer that predicts a binary response that indicates whether a DHS exhibits IR or not. The convolutional filters of the first layer are used to extract position weight matrices (PWMs) that are searched against a database of known TFs