Modeling Single-Cell ATAC-Seq Data Based on Contrastive Learning

With the advance of single-cell assay for transposase-accessible chromatin sequencing technologies (scATAC-seq), it is able to assess the accessibility of single-cell chromatin and gain insights into the process of gene regulation. However, the scATAC data contains distinct characteristics such as sparsity and high dimensionality, which often pose challenges in the downstream analysis. In this paper, we introduce a contrastive learning method (SCCL) for modeling scATAC data. The SCCL designs two distinct encoders to extract local and global features from the original data, respectively. In addition, an improved contrastive learning method is utilized to reduce the redundancy of the feature. Further, the local and global features are fused to obtain reliable features. Finally, the decode is used to generate binary accessibility. We conduct the experiment on various real datasets, and the results demonstrate its superiority over other state-of-the-art methods in cell cluster and transcription factor activity inference.