scMCG: A Method for Analyzing scATAC-seq Data Based on Contrastive Learning and Generative Adversarial Network

The development of single-cell Assay for Transposase-Accessible Chromatin sequencing technologies (scATAC-seq) has greatly advanced the study of cell heterogeneity in the epigenetic landscape. Numerous studies have leveraged scATAC data to explore deeper gene regulatory relationships. In this paper, we introduce a method (scMCG) for analyzing scATAC-seq data which employs a generative adversarial strategy and contrastive learning. First, scMCG utilizes two distinct encoders for contrastive learning to reduce feature redundancy in scATAC data. Then it uses a generative adversarial network(GAN) to reconstruct latent embeddings. It enables the latent embeddings to better represent complex high-dimensional data. Finally, the decoder is used to generate binary accessibility. We conduct experiment on multiple scATAC datasets, and the results demonstrate that the model achieves excellent performance in tasks such as cell clustering and transcription factor activity influence.