matchSCore: Matching Single-Cell Phenotypes Across Tools and Experiments

Single-cell transcriptomics allows the identification of cellular types, subtypes and states through cell clustering. In this process, similar cells are grouped before determining co-expressed marker genes for phenotype inference. The performance of computational tools is directly associated to their marker identification accuracy, but the lack of an optimal solution challenges a systematic method comparison. Moreover, phenotypes from different studies are challenging to integrate, due to varying resolution, methodology and experimental design. In this work we introduce matchSCore (https://github.com/elimereu/matchSCore), a measure to fastly match cell populations across tools, experiments and technologies. We compared 14 computational methods and evaluated their accuracy in clustering and gene marker identification in simulated data sets. Further, we used matchSCore to project cell type identities across mouse or human cell atlas projects. Despite originated from different technologies, cell populations could be matched across datasets, allowing the assignment of clusters to reference maps and their annotation. We produced single-cell transcriptome profiles for pancreas and bladder tissue and projected our and publically available cell clusters onto a human or mouse organ reference atlas, respectively.

单细胞转录组学（single-cell transcriptomics）可通过细胞聚类实现细胞类型、亚型及状态的鉴定。该流程中，先对相似细胞进行聚类分组，随后确定共表达标记基因以完成表型推断。各类计算工具的性能与其标记基因鉴定准确率直接相关，但由于缺乏最优解决方案，系统性的方法对比面临挑战。此外，受分辨率、研究方法及实验设计差异的影响，不同研究得到的表型难以整合。本研究中我们推出了matchSCore（https://github.com/elimereu/matchSCore），一种可快速匹配不同工具、实验及技术下细胞群的量化指标。我们对14种计算方法进行了对比，并在模拟数据集上评估了它们在细胞聚类与基因标记鉴定中的准确率。进一步地，我们利用matchSCore实现了小鼠或人类细胞图谱项目间细胞类型标识的跨数据集映射。尽管不同数据集源自不同技术，仍可实现细胞群间的匹配，从而完成聚类簇向参考图谱的分配与注释。我们分别生成了胰腺与膀胱组织的单细胞转录组图谱，并将本研究及公开可用的细胞聚类簇分别映射至人类与小鼠器官参考图谱中。