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

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution gene expression profiling, revealing cellular heterogeneity and rare cell populations.
• scRNA-seq data's high dimensionality, sparsity, and noise present challenges for traditional clustering methods in accurately identifying cell types and states.
• Conventional algorithms struggle with the complex geometry of gene expression landscapes and may not account for non-linear manifold structures.

Purpose of the Study:

• To develop and evaluate advanced unsupervised clustering algorithms for scRNA-seq data.
• To improve the accuracy of cell type and state delineation from complex transcriptomic profiles.
• To address the limitations of existing clustering methods in handling scRNA-seq data characteristics.

Main Methods:

• Proposed a random sampling-based consensus clustering framework.
• Integrated manifold-based spectral clustering for predicting the optimal number of clusters.
• Incorporated geodesic distance and discrete weighting measures into a similarity kernel to enhance clustering performance.

Main Results:

• The proposed framework demonstrated improved clustering performance on real-world scRNA-seq datasets.
• Geodesic distance-based similarity kernel with discrete weighting showed superior results on datasets with discrete variations.
• The methods effectively addressed challenges posed by dimensionality, sparsity, and noise in scRNA-seq data.

Conclusions:

• The developed clustering framework offers a robust approach for analyzing scRNA-seq data.
• Geodesic distance and discrete weighting are effective strategies for enhancing clustering accuracy in specific scRNA-seq data types.
• This work contributes to more precise identification of cellular heterogeneity and subpopulations using transcriptomic data.