DeepMILO: a deep learning approach to predict the impact of non-coding sequence variants on 3D chromatin structure

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Non-coding variants can influence disease by altering 3D genome organization.
• CTCF/cohesin-mediated insulator loops are crucial for genome structure and function.

Purpose of the Study:

• To develop DeepMILO, a deep learning method for predicting variant effects on insulator loops.
• To identify disrupted insulator loops in cancer patients using whole-genome sequencing data.

Main Methods:

• Deep learning model (DeepMILO) developed to predict variant impact on CTCF/cohesin-mediated insulator loops.
• Application of DeepMILO to whole-genome sequences from 1834 patients across twelve cancer types.

Main Results:

• Identified 672 insulator loops disrupted in at least 10% of patients.
• Found mutations at loop anchors associated with BCL2 and MYC upregulation in malignant lymphoma.

Conclusions:

• DeepMILO effectively predicts variant effects on insulator loops.
• Disrupted insulator loops may represent a novel mechanism for cancer driver gene dysregulation, particularly BCL2 and MYC in lymphoma.