Reference Vector-guided Evolutionary Algorithm for cluster analysis of single-cell transcriptomes

Abstract

Background and objective: Single-cell RNA-sequencing (scRNA-seq) has revolutionized transcriptomic studies by providing detailed insights into gene expression profiles at the single-cell level. This technology allows researchers to capture expression patterns of thousands of genes across hundreds or thousands of individual cells. Clustering is a crucial step in the analysis of scRNA-seq data, since it enables the identification of distinct cell populations based on their transcriptomic profiles and serves as a foundation for downstream analysis. Given that clustering scRNA-seq data is a challenging task that involves different conflicting objectives, our goal is to tackle it from a multi-objective optimization perspective.

Methods: This study proposes a Reference Vector-guided Evolutionary Algorithm for Cluster Analysis of Single-cell Transcriptomes (RVEA-CAST) to address the clustering task as a multi-objective optimization problem. Our approach considers three objectives to optimize: clustering deviation, clustering compactness, and the Davies-Bouldin index. The algorithmic design of RVEA-CAST incorporates three problem-aware mutation operators specifically designed to improve each objective, which are orchestrated under a multi-objective search engine based on the use of reference vectors.

Results: RVEA-CAST is evaluated on ten real scRNA-seq datasets using standard clustering evaluation metrics, such as Normalized Mutual Information (NMI) and Adjusted Rand Index (ARI). The attained results reveal the improved performance and robustness of the proposed approach compared to other previously proposed methods. Specifically, statistically significant improvements of up to 66.7% and 261.5% were achieved for NMI and ARI, respectively. Furthermore, the analysis of differentially expressed genes in the predicted and real clusters showcased greater agreement of our solutions with actual cell populations, underscoring the biological relevance of our approach.

Conclusions: The results highlight that RVEA-CAST is an effective and versatile approach for clustering scRNA-seq data, outperforming existing methods across diverse biological scenarios in both widely used clustering evaluation metrics and biological relevance.

Keywords: Cluster analysis; Multi-objective optimization; Reference Vector-guided Evolutionary Algorithm; ScRNA-seq; Single-cell transcriptome.