Hi-Cformer enables multiscale chromatin contact map modeling for single-cell Hi-C data analysis

Description

Xiaoqing Wu, Zian Wang, Rui Jiang, and Xiaoyang Chen teach a transformer to read single-cell chromatin contact maps, the sparse fingerprints of how one cell folds its genome. Single-cell Hi-C data are desperately sparse and pile their signal near the diagonal, defeating standard models. Real structure spans many scales at once.

Their 2026 paper, Hi-Cformer, attends over multiscale blocks of the contact map jointly, catching dependencies across regions and scales in one pass. The learned cell embeddings separate cell types more cleanly than rival methods. The model also fills in missing contacts, recovering domain-like boundaries and compartment patterns tied to cell identity. Those embeddings then annotate cell types robustly across data sets, a practical prize for single-cell atlases.

Catalog record

Source: https://europepmc.org/articles/PMC13524026