scICER：面向整合单细胞分析的分群稳定性评估工具
scICER: clustering stability assessment for integrated single-cell analyses
scICER 是单细胞不一致性聚类估计（scICE）分析的 R/Seurat 实现，可直接在 Seurat 对象中用户提供的图上评估分群稳定性。相比原始 Julia 实现，scICER 增加了生态原生输入输出、固定分辨率评估、极小簇的可选处理和面向大型对象的实用流程。在 48 个基准数据集及整合分析中，scICER 复现了参考 scICE 的行为，并支持在 Harmony 或 scVI 校正后进行稳定性评估。工具以 R 实现并免费公开，另有功能相近的 Python 实现 scICEpy 可用。
来源：Bioinformatics Advances · doi.org
Motivation
Clustering is essential in single-cell and spatial omics for identifying cell types and states, but stochastic algorithms such as Leiden can reduce cluster stability and compromise annotation reliability. We present scICER, an R/Seurat implementation of single-cell Inconsistency Clustering Estimator (scICE)-style inconsistency analysis that evaluate clustering stability on user-provided graphs stored in Seurat objects from standard or integrated workflows.
Results
Compared with the original Julia implementation, the scICER adds ecosystem-native input/output, fixed-resolution evaluation, optional handling of very small clusters, and practical workflows for large objects. Across 48 benchmark datasets and integrated analyses, scICER reproduced reference scICE behavior while enabling stability assessment after Harmony or scVI correction.
Availability and implementation
scICER is implemented in R and is freely available at https://github.com/ATPs/scICER. A companion Python implementation with similar functionality, scICEpy, is available at https://github.com/ATPs/scICEpy.