Preprocessing of Single Cell RNA Sequencing Data Using Correlated Clustering and Projection

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity, crucial for understanding cell communication, differentiation, and gene expression.
• Analyzing scRNA-seq data is challenging due to data sparsity and high dimensionality, necessitating effective dimensionality reduction and feature selection.

Purpose of the Study:

• To introduce Correlated Clustering and Projection (CCP), a novel data-domain dimensionality reduction technique for scRNA-seq data.
• To demonstrate CCP's advantages over Principal Component Analysis (PCA) for high-dimensional clustering and classification tasks.
• To present the Residue-Similarity index (RSI) and R-S plot as new tools for evaluating and visualizing clustering and classification performance.

Main Methods:

• Developed CCP, a method that projects clusters of similar genes into 'supergenes' based on accumulated pairwise nonlinear gene-gene correlations.
• Evaluated CCP using 14 benchmark scRNA-seq datasets.
• Introduced the Residue-Similarity index (RSI) for performance assessment and the R-S plot for visualization.

Main Results:

• CCP demonstrated significant advantages over PCA for clustering and classification on intrinsically high-dimensional data.
• The RSI metric showed a strong correlation with accuracy, even without true labels.
• The R-S plot offered a viable alternative to UMAP and t-SNE for visualizing data with numerous cell types.

Conclusions:

• CCP is an effective dimensionality reduction method for scRNA-seq data, outperforming PCA in specific tasks.
• RSI and R-S plots provide valuable, label-independent tools for assessing and visualizing scRNA-seq analysis results.
• CCP and its associated tools offer new avenues for exploring cellular heterogeneity and complex biological systems.