Evaluation of deep learning-based feature selection for single-cell RNA sequencing data analysis.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Feature selection is crucial for single-cell RNA sequencing (scRNA-seq) data analysis, aiding dimension reduction and downstream tasks like gene marker identification and cell type classification.
• Traditional methods often rely on differential distribution, while newer deep learning approaches determine gene importance using neural networks.

Purpose of the Study:

• To explore the effectiveness of various deep learning-based feature selection methods for scRNA-seq data.
• To compare deep learning methods against traditional approaches in terms of performance and efficiency.

Main Methods:

• Utilized scRNA-seq datasets sampled from the Tabula Muris and Tabula Sapiens atlases.
• Evaluated traditional and deep learning feature selection methods on metrics including cell type classification accuracy, feature selection reproducibility, diversity, and computational time.

Main Results:

• Deep learning methods show utility in scRNA-seq feature selection.
• Performance was assessed across diverse datasets, providing insights into method robustness and efficiency.

Conclusions:

• This research serves as a reference for applying and developing deep learning-based feature selection in single-cell omics.
• Highlights the potential of deep learning for advancing scRNA-seq data analysis and interpretation.