Improved Spectral Clustering Method for Identifying Cell Types from Single-Cell Data

Abstract

With the development of single-cell RNA sequencing (scRNA-seq) technology, characterizing heterogeneity at the cellular level has become a new area of computational biology research. However, the infiltration of different types of cells and the high variability in gene expression complicate classification of cell types. In this study, we propose an improved spectral clustering method for clustering single-cell data that avoid the overfitting issue and consider both similarity and dissimilarity, motivated by the observation that same type cells have similar gene expression patterns, but different types of cells produce dissimilar gene expression patterns. To evaluate the performance of the proposed spectral clustering method, we compare it with the traditional spectral clustering method in recognizing cell types on various real scRNA-seq data. The results show that taking intercellular dissimilarity into account can effectively achieve high accuracy and robustness and that our method outperforms the traditional spectral clustering methods in grouping cells that belong to the same cell types.