February 22, the research team led by Shengquan Chen from the School of Mathematical Sciences at Nankai University published a paper entitled scCASE: accurate and interpretable enhancement for single-cell chromatin accessibility sequencing data in the famous international journal Nature Communications. This paper proposes scCASE, a single-cell chromatin accessibility sequencing (scCAS) data enhancement method based on non-negative matrix factorization which incorporates an iteratively updating cell-to-cell similarity matrix. In this way, epigenetic signals of similar cells are effectively integrated to reduce the noise level. 

The graphical illustration of scCASE

ScCASE preprocesses the scCAS count matrix by filtering the peaks that are accessible in less than 1% of cells, and reweighting the count matrix based on term frequency - inverse document frequency (TF-IDF). Then, scCASE factorizes the count matrix into a projection matrix and a cell embedding matrix. Since similar cells usually have similar chromatin accessibility patterns, scCASE additionally introduces an iteratively updating cell-to-cell similarity matrix to achieve data imputation and enhancement. Finally, the enhanced data are reconstructed by minimizing loss function through gradient descent. 

Through comprehensive experiments on multiple datasets, the paper systemically demonstrates the superior performance of scCASE over other methods in data enhancement, downstream analysis, robustness, etc. The enhanced single-cell chromatin accessibility data can effectively capture cellular heterogeneity signals and facilitate downstream analysis such as cell clustering and visualization. Through extensive tissue-specific expression enrichment, biological function enrichment, and heritability analysis, the paper shows that scCASE can provide valuable biological insights into cell subpopulations. Finally, the paper provides multiple extended variations and demonstrates their potential in sequencing depth correction, batch effect correction, and weakly supervised learning with reference data.

Nankai University is the first completion unit and communication unit for this work. Associate Professor Shengquan Chen from Nankai University is the corresponding author, and Songming Tang, a 2023 graduate student from the School of Mathematical Sciences at Nankai University, is the first author.

https://www.nature.com/articles/s41467-024-46045-w

（Edited and translated by Nankai News Team.)