Researchers introduced a few-shot AI method, scCAPReSE, that detects large-scale genomic rearrangements directly from single-cell Hi-C data. The framework leverages pre-trained image-model representations and reports over 90% classification accuracy, while also surfacing cell-to-cell variability in 3D cancer genome organization. The approach is aimed at resolving rearrangement detection in settings where labeled training examples are limited. Single-cell Hi-C is a technology that maps chromatin contacts at the single-cell level, generating interaction data that can be converted into images or feature maps. If validated beyond benchmark settings, the method could reduce barriers to translating 3D genome alterations into actionable biomarkers, particularly in heterogenous tumor samples where existing approaches struggle.