Separating measurement and expression models clarifies confusion in single-cell RNA sequencing analysis

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) datasets frequently exhibit a high proportion of zero counts.
• Inconsistent terminology like 'dropout' and 'missing data' complicates interpretation of these zeros.
• Existing methods for analyzing scRNA-seq data often use complex models without clear justification.

Purpose of the Study:

• To clarify and standardize terminology surrounding zero counts in scRNA-seq data.
• To propose a simplified modeling framework for analyzing scRNA-seq data.
• To provide a unified perspective on existing scRNA-seq analysis methods.

Main Methods:

• Distinguishing between true gene expression levels and technical measurement error in scRNA-seq counts.
• Advocating for the use of a Poisson measurement model as a foundational approach for method development.
• Re-evaluating existing scRNA-seq methods within the proposed Poisson framework.

Main Results:

• Observed scRNA-seq counts are a combination of true expression and measurement error.
• A Poisson measurement model provides a simple yet consistent starting point for analyzing scRNA-seq data.
• Existing methods can be understood based on their differing assumptions about expression variation within this framework.

Conclusions:

• Clarifying the distinction between biological signal and technical noise is crucial for accurate scRNA-seq analysis.
• A Poisson model offers a robust and interpretable foundation for developing new scRNA-seq analysis tools.
• This perspective aids in addressing fundamental biological questions, such as the distribution of gene expression levels across cells.