|
Abstract: |
Single-cell sequencing (scSeq) technologies are revolutionizing biomedical research and clinical practice by enabling the comprehensive characterization of cells from complex tissues. scSeq can identify cell types, states, and lineages of different cell subpopulations associated with diseases in a heterogeneous tissue ecosystem, providing a breakthrough opportunity to understand biological and clinical mechanisms. However, interpreting complex single-cell data from highly heterogeneous cell populations remains challenging. Identifying cell subpopulations that drive phenotypes, such as tumor metastasis, treatment resistance, and survival outcome, is of indispensable importance since it will facilitate cell-type targeted therapies and prognostic biomarker discovery. Clustering-based methods depend on the subjective clustering step and are often suboptimal since unsupervised clustering may not capture phenotype-specific subpopulations.
To overcome this challenge, we recently developed a novel framework called SCISSOR, which identifies phenotype-associated subpopulations from single-cell assays by leveraging the wealth of phenotype and bulk-omics datasets (Sun et al., Nature Biotechnology, 2022). Thus, this tool paves the way for employing widely available bulk phenotype information to unlock the disease-relevant subpopulations for cellular target discoveries from scSeq data. Moreover, an increasing number of single-cell experiments are designed to profile multiple samples from different conditions, such as treatment resistance versus responder groups. Identifying subpopulations unique to each phenotype will improve phenotype-specific gene signal detection to facilitate reliable downstream analysis. Since clustering-based methods are often suboptimal, we propose to use a novel 'learning with rejection' platform to learn high confidence phenotype-enriched subpopulations from millions of cells (Ren et al., Nature Machine Intelligence, 2023) |
