Fig. 1

Overall approach for the identification of insulator loops and genes affected by mutations in patients. Anchors of insulator loops and non-loops are used to train DeepMILO to learn sequence features of insulator loops. The model is then applied to patients’ mutation data (i.e., structural variants, single nucleotide variants, and indels) to detect insulator loop changes associated with mutations. Insulator loops can be the set of default insulator loops from 4 cell lines GM12878, K562, MCF7, and Hela or a set of cell type-specific insulator loops provided by users