HOMC: A Hierarchical Clustering Algorithm Based on Optimal Low Rank Matrix Completion for Single Cell Analysis

The tremendous development of single-cell RNA sequencing (scRNA-seq) technology offers the promise of addressing cellular heterogeneity problem which cannot be addressed with bulk sequencing technologies. However, scRNA-seq data is noisy and sparse due to the dropout events. In this study, we focused on cellular heterogeneity problem and proposed a hierarchical clustering algorithm based on optimal low rank matrix completion (HOMC). We first applied nonnegative matrix factorization for determining optimal low rank approximation for the original scRNA-seq data. Then we performed hierarchical clustering based on correlation-based distance for grouping those imputed data points, and optimal number of clusters can be determined by integrating three classical measures. Experimental results have showed that HOMC is capable of distinguishing cellular differences and the clustering performance is superior to other state-of-the-art methods.