A clustering method for small scRNA-seq data based on subspace and weighted distance.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Unsupervised cell type identification is crucial for single-cell RNA sequencing (scRNA-seq) research.
• Conventional similarity measures struggle with scRNA-seq data's high dimensionality, noise, and dropout rates.
• Accurate cell clustering is essential for interpreting scRNA-seq results.

Purpose of the Study:

• To develop a robust clustering method for small scRNA-seq datasets.
• To address the limitations of existing similarity measures in scRNA-seq analysis.
• To improve the accuracy and partitioning ability of cell group identification.

Main Methods:

• Proposed a novel clustering method named Subspace and Weighted Distance (SSWD).
• Introduced a new distance metric combining Euclidean and Pearson distances with a weighting strategy.
• Utilized the relative Calinski-Harabasz (CH) index for estimating optimal cluster numbers.

Main Results:

• SSWD demonstrated superior clustering accuracy compared to seven prevailing methods.
• The method showed enhanced partitioning ability for distinguishing cell groups.
• Experiments were conducted on eight publicly available scRNA-seq datasets.

Conclusions:

• SSWD offers a more accurate and effective approach for scRNA-seq data clustering.
• The proposed method successfully addresses challenges posed by noisy and high-dimensional single-cell data.
• SSWD provides a valuable tool for cell type identification in scRNA-seq research.