G4SCOPE identifies sequence features and epigenetic contexts of chromatin G-quadruplexes

Researchers have developed a cutting-edge deep learning framework called G4SCOPE to unravel the intricate dynamics of chromatin-associated G-quadruplex (G4) structures. This innovative tool models G4 occupancy and reconstructs high-resolution CUT&Tag profiles solely from primary DNA sequences, overcoming the challenges of predicting endogenously folded G4s.

Trained on genotype-stratified G4 enriched cut&tag landscapes across various human breast epithelial cell lines, G4SCOPE integrates convolutional motif learning with transformer self-attention to decode long-range sequence contexts. The model effectively distinguishes chromatin-associated G4s from GC-matched loci with intrinsic folding potential, showcasing its robustness and versatility.

Remarkably, genotype-resolved mapping using G4SCOPE reveals a significant divergence in genome maintenance states. BRCA1 deficiency leads to a global inflation of chromatin-embedded G4 density, while BRCA2 loss is associated with a systemic, chromosome-wide structural depletion. The model's unsupervised clustering of latent sequence representations uncovers distinct biological G4 subclasses occupying unique regulatory class genomic environments, which are characterized by unique 5-methylcytosine and 5-hydroxymethylcytosine landscapes.

In summary, G4SCOPE decodes the primary sequence grammar governing cell-based G4 occupancy, demonstrating that sequence-defined G4 topologies harbor intrinsic and distinct local epigenetic features. This powerful framework paves the way for a deeper understanding of G4 dynamics and their implications in cancer-associated epigenetic remodeling.