An Unsupervised Tool for Single-Cell Clustering Result Evaluation

• Activity Number: 81 - Contributed Poster Presentations: Section on Statistics in Epidemiology
• Type: Contributed
• Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Jiyuan Fang* and Jichun Xie and Qijing Li and Cliburn Chan
• Companies: Duke University and Duke University and Duke University and Duke University
• Keywords: single-cell; clustering; unsupervised

Abstract

Single-cell RNA-sequencing (scRNA-seq) technology allows us to explore the cellular heterogeneity, and most analyses of scRNA-seq data start from the clustering of cells. Hence, clustering accuracy is crucial for downstream analysis. Although hundreds of clustering methods have been developed, few tools are available to evaluate the clustering accuracy. Commonly used tools such as the adjusted Rand index requires knowledge of the true cell type labels, which are often unavailable in the data. Here, we develop an unsupervised tool to evaluate the data-specific clustering results, which does not require knowledge of the true cell type labels. The tool will generate a clustering accuracy score (CAS). We demonstrate with the simulated and benchmarked real data that the CAS is consistent with supervised metrics such as the adjusted Rand index. From CAS, we infer the misclassified percentage of the cell population. In addition, our tool can help to choose optimal tuning parameters for any clustering methods.