A Novel Single-Cell RNA Sequencing Data Feature Extraction Method Based on Gene Function Analysis and Its...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cell heterogeneity and identifying new cell subtypes.
• Existing scRNA-seq clustering methods struggle with data noise, sparsity, and poor annotation, often neglecting gene functions and interactions.
• Feature extraction is key to improving the accuracy and interpretability of scRNA-seq data analysis.

Purpose of the Study:

• To develop and evaluate a novel feature extraction method, FEGFS, for scRNA-seq data analysis.
• To improve cell clustering by incorporating gene function information.
• To demonstrate the effectiveness of FEGFS compared to state-of-the-art methods.

Main Methods:

• Derived functional gene sets from Gene Ontology (GO) terms, reducing redundancy via semantic similarity and gene repetitive rate.
• Applied kernel principal component analysis (KPCA) for feature selection within each non-redundant functional gene set.
• Combined selected features for subsequent clustering analysis using agglomerative hierarchical clustering.

Main Results:

• FEGFS significantly outperformed existing methods on small scRNA-seq datasets (Pollen, Goolam) across multiple evaluation metrics (ARI, NMI, HOM, COM).
• On large datasets (Klein, Zeisel), FEGFS showed competitive performance, outperforming most methods and demonstrating its scalability.
• The study also highlighted the utility of CMF-Impute for reconstructing correlations and inferring cell lineage trajectories, with applications in identifying glioma-related cell clusters and marker genes.

Conclusions:

• FEGFS offers a robust approach to feature extraction for scRNA-seq data, effectively utilizing gene ontology information.
• The method enhances cell clustering accuracy, particularly for smaller datasets, and shows promise for larger, more complex biological data.
• FEGFS and associated imputation methods provide valuable tools for dissecting cellular heterogeneity and identifying disease-specific markers, as exemplified in glioma research.