Wasserstein CycleGAN for Single-Cell Rna- Seq Data Generation Using Cross-Modality Translation

Abstract

Single-nucleus RNA sequencing (snRNA-seq) provides insights into gene expression in complex tissues but suffers from lower resolution compared to single-cell RNA sequencing (scRNA-seq). To bridge this gap, we propose scWC-GAN, a Wasserstein CycleGAN-based model that translates snRNA-seq data into high-resolution scRNA-seq profiles. Our method leverages Earth Mover’s Distance (EMD) for cycle consistency and a latent feature-preserving generator to capture transcriptomic structures better. Through extensive evaluation, scWC-GAN outperforms baseline models in FID score and SSIM, demonstrating its ability to generate biologically meaningful data. While challenges remain in fine-grained cell-type resolution, our results suggest scWC-GAN as a promising tool for cross-modality single-cell data translation, enhancing downstream analysis in genomics.