GTestimate: improving relative gene expression estimation in scRNA-seq using the Good-Turing estimator.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is susceptible to technical variations due to complex experiments and shallow sequencing.
• Conventional normalization methods often use suboptimal maximum likelihood estimators for relative gene expression per cell.
• These limitations can impact the accuracy of downstream scRNA-seq data analysis.

Purpose of the Study:

• To introduce GTestimate, a novel normalization method for scRNA-seq data.
• To improve the estimation of relative gene expression by accounting for unobserved genes.
• To enhance the accuracy of cell-cell distance estimations in scRNA-seq data.

Main Methods:

• Developed GTestimate, a normalization method utilizing the Good-Turing estimator.
• Introduced cell-targeted PCR amplification sequencing (cta-seq) for ultra-deep single-cell sequencing.
• Validated GTestimate using cta-seq data and explored its compatibility with Seurat workflows on four datasets.

Main Results:

• GTestimate demonstrates improved relative gene expression estimation compared to conventional methods.
• The Good-Turing estimator enhances cell-cell distance estimation accuracy.
• GTestimate integration with Seurat workflows improved downstream analysis results across example datasets.

Conclusions:

• Employing a more suitable estimator, like Good-Turing, significantly enhances scRNA-seq normalization.
• GTestimate offers a user-friendly R-package compatible with various workflows, facilitating widespread adoption.
• Improved normalization has substantial implications for the reliability and interpretation of scRNA-seq downstream results.