Inferring CTCF binding patterns and anchored loops across human tissues and cell types

Abstract

How CTCF recognizes insulators to exert chromosome barrier or enhancer blocking effects remains to be interrogated. Despite many computational tools were developed to predict CTCF-mediated loops qualitatively or quantitatively, few could specially evaluate the insulative potential of DNA sequence at CTCF binding sites (CBSs) and how it affects chromatin loop formation. Here, we developed a deep learning model, DeepAnchor, to precisely predict the insulative potential of CBS. By incorporating base-wise genomic/epigenomic features, we revealed distinct chromatin and sequence features for CTCF-mediated insulation at a high resolution, such as two sequence motifs flanking the core CTCF motif only at positive CTCF insulators. Besides, we leveraged the predicted insulator score to optimize the loop extrusion model and achieved the best performance in predicting CTCF-anchored loops. We established a compendium of context-specific CTCF-anchored loops across 52 human tissue/cell types and found that genomic disruption of CTCF-anchored loops may represent a general causal mechanism of disease pathogenesis. These computational models, together with the established resource, could facilitate the mechanistic research on how the CTCF-mediated insulation shapes context-specific gene regulation in cell development and disease progression.

Stage-resolved chromatin conformation dynamics during human meiosis

“…Computational methods that employ machine learning strategies to predict chromatin looping (Kai et al 2018;Matthews and Waxman 2018;Xi and Beer 2021;Xu et al 2023) are a timely substitute when experimental data are hard to come by. Machine learning models can be trained on experimental data, such as CTCF-binding, gene expression and other epigenetic marks in combination with a set of "true" loops derived from Hi-C or ChiA-Pet datasets.…”