scTopoDEC: topological deep embedded clustering for single-cell RNA-seq data with persistent homology

Clustering remains a challenging task in scRNA-seq analysis despite the development of numerous computational methods. Here, we introduce scTopoDEC (single-cell topological deep embedded clustering), a deep learning-based clustering method that incorporates persistent homology to improve clustering performance by preserving topological information in single-cell data. We evaluated scTopoDEC on simulated and real-world benchmark datasets and demonstrated that it achieved superior clustering performance compared with existing methods.

Moreover, we applied scTopoDEC to a case study dataset of chronic lymphocytic leukaemia (CLL), where the model identified an RT-like tumour cell population with molecular characteristics associated with Richter's transformation (RT). These findings highlight the potential of scTopoDEC as a promising tool for single-cell clustering and the identification of biologically distinct cellular populations.