Dimension Reduction and Clustering Models for Single-Cell RNA Sequencing Data: A Comparative Study.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) generates vast transcriptome data, advancing cellular heterogeneity studies.
• scRNA-seq data is high-dimensional, noisy, and sparse, challenging traditional clustering methods.
• Dimension reduction techniques are crucial for effective analysis of scRNA-seq data.

Purpose of the Study:

• To comprehensively review and evaluate classical dimension reduction methods for scRNA-seq data clustering.
• To identify optimal strategies for improving clustering performance on complex scRNA-seq datasets.
• To establish a reliable research routine for scRNA-seq data analysis.

Main Methods:

• Conducted a comprehensive review of four dimension reduction methods and five clustering models.
• Performed four experiments on two large scRNA-seq datasets using 20 combined models.
• Evaluated the impact of feature selection and feature extraction on clustering performance.

Main Results:

• Feature selection positively impacted high-dimensional, sparse scRNA-seq data.
• Feature extraction methods generally improved clustering performance.
• Principal Component Analysis (PCA) demonstrated steadier performance than other feature extraction methods.
• Independent Component Analysis (ICA) excelled in small compressed feature spaces but was not ideal for fuzzy C-means clustering.
• K-means clustering combined with feature extraction yielded good results.

Conclusions:

• Dimension reduction, particularly PCA and feature selection, is vital for effective scRNA-seq data clustering.
• The choice of dimension reduction and clustering method impacts analytical outcomes.
• Optimized combinations of feature extraction and clustering, like K-means, enhance scRNA-seq data analysis.