ClustoCell reveals cell states and their markers from single-cell transcriptomes

A new reference-free method named ClustoCell has been introduced to accurately identify cell types and states from single-cell transcriptome data. This innovative approach addresses the limitations of existing techniques that are often sensitive to biological variability, preprocessing choices, and reference selection. ClustoCell constructs cell-cell similarity graphs by categorizing gene expression of each cell into high and medium tiers, focusing on intrinsic expression patterns rather than overall variance.

In a comprehensive evaluation involving 450 diverse datasets containing over 24 million cells, ClustoCell demonstrated exceptional performance, accurately recovering expert annotations with an impressive 92% concordance. When compared to the most advanced methods currently available, ClustoCell proved to be superior in identifying stable and coherent cell types and states, preventing the unwarranted partitioning of closely related cells, and enhancing the discovery of cell type-specific markers.

One of the most compelling features of ClustoCell is its ability to discern rare and transitional cell states solely based on transcriptional structure. It effectively distinguishes malignant cells from non-malignant ones and refines existing expert cell annotations. Furthermore, when applied to immunotherapy datasets, ClustoCell revealed intricate pre-treatment immune circuits that link T cell states to PD-1 responsiveness, specifically within a tumor-type-specific context.

ClustoCell stands out for its interpretability and scalability, offering a robust foundation for single-cell analysis and translational profiling. By resolving complex cellular architectures and providing valuable insights into cellular states, ClustoCell promises to revolutionize the field of single-cell RNA-sequencing and contribute significantly to our understanding of cellular biology.