Consensus spectral clustering with weighted similarity functions for single-cell RNA sequencing data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution gene expression insights.
• scRNA-seq data presents challenges like high dimensionality, sparsity, and noise.
• Traditional clustering methods struggle with complex scRNA-seq data structures.

Purpose of the Study:

• To improve unsupervised clustering performance on single-cell RNA sequencing datasets.
• To address limitations of conventional clustering algorithms in capturing complex gene expression landscapes.
• To develop a robust framework for accurate cell type and state delineation.

Main Methods:

• Proposed a random sampling-based consensus clustering framework.
• Integrated manifold-based spectral clustering for cluster number prediction.
• Incorporated geodesic distance and discrete weighting measures into similarity kernels.

Main Results:

• The proposed geodesic distance-based similarity kernel with discrete weighting demonstrated superior performance.
• This approach excelled on scRNA-seq datasets exhibiting discrete variations.
• Effectively enhanced clustering accuracy for complex biological data.

Conclusions:

• The developed framework significantly improves unsupervised clustering of scRNA-seq data.
• Geodesic distance and discrete weighting are crucial for handling data variations.
• This method advances the accurate identification of cellular heterogeneity in transcriptomic profiles.