The Poisson distribution model fits UMI-based single-cell RNA-sequencing data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) data modeling is challenged by high zero counts and heterogeneity.
• Existing models often aggregate data at gene or cell levels, leading to accuracy loss.
• Improved scRNA-seq data modeling can significantly benefit downstream analyses.

Purpose of the Study:

• To develop a novel method for modeling scRNA-seq data that addresses limitations of existing aggregation-based approaches.
• To introduce a new data representation, Departures from an independent Poisson distribution (DIPD), for enhanced cell clustering.
• To uncover novel cell subtypes missed by conventional methods.

Main Methods:

• Proposed an independent Poisson distribution (IPD) model for individual entries in the scRNA-seq data matrix.
• Developed a novel data representation, DIPD, to capture per-gene-per-cell heterogeneity for clustering.
• Utilized crafted experiments and real data for validation of the DIPD-based clustering pipeline.

Main Results:

• The IPD model naturally handles the high number of zeros in scRNA-seq data.
• DIPD effectively captures intrinsic heterogeneity generated by cell clusters.
• Experiments demonstrated that DIPD can uncover novel cell subtypes missed by conventional methods.

Conclusions:

• The DIPD method offers advantages such as no need for prior feature selection or hyperparameter tuning.
• The approach is flexible and can enhance other methods like Seurat.
• A new DIPD-based clustering pipeline is implemented in the R package scpoisson, validated using crafted experiments.