scLENS: data-driven signal detection for unbiased scRNA-seq data analysis.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) data is high-dimensional and noisy, limiting biological discovery.
• Existing dimensionality reduction tools often require manual parameter tuning, introducing user bias and potential signal distortion.
• Log normalization, a common preprocessing step, can inadvertently distort biological signals in scRNA-seq data.

Purpose of the Study:

• To develop a novel dimensionality reduction tool, scLENS, for scRNA-seq data that addresses signal distortion and manual input.
• To improve the accuracy of biological insight extraction from complex and noisy scRNA-seq datasets.
• To provide a user-friendly, automated solution for scRNA-seq data analysis.

Main Methods:

• Developed scLENS, a dimensionality reduction technique that uniformizes cell vector lengths using L2 normalization to correct log normalization-induced signal distortion.
• Implemented random matrix theory-based noise filtering for data-driven determination of signal dimensions.
• Incorporated a signal robustness test to refine the threshold for signal dimensions, minimizing user bias.

Main Results:

• scLENS effectively mitigates signal distortion caused by log normalization.
• The tool accurately identifies signal dimensions using data-driven methods, bypassing manual tuning.
• scLENS demonstrates superior performance compared to 11 existing dimensionality reduction tools, especially on sparse and variable scRNA-seq datasets.

Conclusions:

• scLENS offers an automated and accurate method for dimensionality reduction in scRNA-seq data.
• The tool enhances the discovery of biological insights by overcoming limitations of current methods.
• scLENS provides a robust solution for analyzing challenging scRNA-seq datasets without user intervention.