PsiNorm: a scalable normalization for single-cell RNA-seq data.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution transcriptome data but presents computational challenges.
• Increased cell throughput in scRNA-seq exacerbates memory and time demands for analysis, including normalization.
• There is a need for accurate and computationally efficient normalization methods for scRNA-seq data.

Purpose of the Study:

• To propose PsiNorm, a novel between-sample normalization method for scRNA-seq data.
• To demonstrate the suitability of the Pareto distribution for modeling scRNA-seq data.
• To develop a scalable and accurate normalization technique that addresses computational bottlenecks.

Main Methods:

• Developed PsiNorm, a normalization method utilizing Pareto distribution parameter estimation.
• Applied PsiNorm to scRNA-seq datasets, particularly those from unique molecular identifier (UMI) platforms.
• Benchmarked PsiNorm against seven other normalization methods using metrics like cluster identification, concordance, and computational resource usage.

Main Results:

• The Pareto distribution effectively models scRNA-seq data, especially UMI-based data.
• PsiNorm demonstrates strong performance in cluster identification and concordance.
• PsiNorm offers a favorable trade-off between accuracy and scalability, outperforming or matching existing methods.
• PsiNorm does not require a reference dataset, enhancing its utility in supervised learning scenarios.

Conclusions:

• PsiNorm is an accurate and highly scalable normalization method for scRNA-seq data.
• The method's independence from reference data makes it versatile for various analytical pipelines, including supervised classification.
• PsiNorm is available within the scone Bioconductor package, facilitating its adoption in the research community.