Abstract
Single-cell RNA-sequencing (scRNA-seq) enables discovery of clinically and biologically interesting populations, but detecting rare cell types is a persistent challenge. Here we introduce Scalpel, a novel technique for extracting interpretable and maximally informative features from single-cell data, enabling population discovery, batch correction, and other downstream analyses at unprecedented resolution. On a collection of cytotoxic T-cells, Scalpel recovers subtle and biologically important populations, including gamma-delta T-cells and MAIT cells, which are invisible to standard pipelines. In multi-batched data, Scalpel effectively removes systemic batch effects, achieving robust and state-of-the-art performance. Unlike other methods, Scalpel is completely unsupervised, human-interpretable, and applicable to both continuous trajectories and clustered data, making it suitable in a wide range of analytic settings.
Competing Interest Statement
The authors have declared no competing interest.