HiCP2GAN: a plug and play foundation model-based GAN for Hi-C enhancement.

HiCP2GAN enhances sparse, low-resolution Hi-C contact maps using a pretrained Vision Transformer discriminator trained on 118 million Hi-C patches across species and cell types.

• Why it matters: High-resolution Hi-C is costly and scarce, and GAN-based enhancement methods often train custom discriminators from scratch, risking unstable training and weak generalization.
• What they did: The plug-and-play framework uses the Hi-C foundation model as a discriminator backbone, finetuning only the first few encoder layers while freezing deeper transformer layers, with a generator-agnostic design.
• The result: Across multiple cell lines, it outperforms traditional GAN-based methods in most settings and improves reconstruction quality and biological reproducibility of existing Hi-C enhancement models in most evaluated cases.

AI summary of the abstract · may contain errors · not medical advice · read the paper.