A comparison of deep learning-based pre-processing and clustering approaches for single-cell RNA sequencing data

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional, sparse, and noisy data, posing challenges for traditional analysis tools.
• The continuous release of scRNA-seq datasets necessitates advanced analytical approaches.
• Deep learning models show significant promise in addressing the complexities of scRNA-seq data.

Purpose of the Study:

• To systematically review popular single-cell RNA sequencing analysis methods and tools based on deep learning models.
• To cover data preprocessing steps including quality control, normalization, data correction, dimensionality reduction, and visualization.
• To evaluate deep learning-based data correction and clustering methods.

Main Methods:

• Systematic literature review of deep learning models for scRNA-seq analysis.
• Quantitative evaluation of deep model-based data correction and clustering on 11 gold standard datasets.
• Analysis of method-specific data preferences and limitations.

Main Results:

• Deep learning models are well-suited for handling the characteristics of scRNA-seq data.
• The review provides a comprehensive overview of deep learning tools for preprocessing and downstream analysis.
• Quantitative evaluations highlight the performance and data preferences of various deep learning methods.

Conclusions:

• Deep learning offers powerful solutions for scRNA-seq data analysis, particularly for preprocessing and clustering.
• Understanding method-specific strengths and limitations is crucial for effective application.
• Guidance is provided for selecting appropriate deep learning tools for scRNA-seq research.