Unveiling patterns: an exploration of machine learning techniques for unsupervised feature selection in single-cell...

Area of Science:

• Single-cell biology
• Computational biology
• Bioinformatics

Background:

• Single-cell technologies generate vast multimodal datasets (genomic, transcriptomic, proteomic, spatial).
• High dimensionality, noise, and computational costs challenge data analysis.
• Traditional feature selection methods (e.g., highly variable gene selection) can introduce bias.

Purpose of the Study:

• To review unsupervised machine learning (ML) techniques for single-cell data analysis.
• To highlight how unsupervised ML minimizes bias and captures complex biological patterns.
• To discuss the potential of these methods for enhancing downstream analyses and biological discovery.

Main Methods:

• Review of unsupervised ML algorithms applicable to single-cell data.
• Discussion of feature selection strategies within unsupervised ML frameworks.
• Exploration of applications in clustering, dimensionality reduction, visualization, and denoising.

Main Results:

• Unsupervised ML identifies informative features without predefined labels, reducing bias.
• These methods can reveal biologically relevant gene modules.
• Successful application enhances various downstream single-cell analyses.

Conclusions:

• Unsupervised ML is crucial for unbiased analysis of complex single-cell data.
• Challenges include data sparsity, parameter tuning, and scalability.
• Future work should focus on multiomic integration, domain knowledge, and scalable algorithms.