scFED: Clustering Identifying Cell Types of scRNA-Seq Data Based on Feature Engineering Denoising

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables disease mechanism investigation at the cellular level.
• Effective clustering and classification of scRNA-seq data rely on high-quality feature sets.
• Computationally intensive and highly expressed genes present challenges for stable feature selection.

Purpose of the Study:

• To introduce scFED, a novel feature-engineered gene selection framework for scRNA-seq data analysis.
• To enhance the accuracy and efficiency of scRNA-seq data clustering and cell type identification.
• To address limitations of existing gene selection methods by reducing noise and incorporating biological knowledge.

Main Methods:

• scFED employs a feature engineering approach to identify robust gene sets.
• It integrates external biological knowledge from the CellMatch database for tissue-specific cellular taxonomy.
• A reconstruction strategy is utilized for noise reduction and amplification of key biological signals.

Main Results:

• scFED demonstrated improved clustering performance on four real-world scRNA-seq datasets.
• The framework effectively reduced data dimensionality.
• Enhanced cell type identification accuracy was observed when scFED was combined with clustering algorithms.
• scFED outperformed existing gene selection techniques in comparative analyses.

Conclusions:

• scFED provides a significant advancement in gene selection for scRNA-seq data.
• The framework offers a robust and efficient method for noise reduction and feature enhancement.
• scFED improves the reliability and interpretability of scRNA-seq data analysis, particularly for disease research.