Dirichlet Process Mixture Model for Correcting Technical Variation in Single-Cell Gene Expression Data

Area of Science:

• Computational Biology
• Genomics
• Data Science

Background:

• Single-cell RNA sequencing (scRNA-seq) enables high-resolution gene expression analysis.
• Technical variations and cell type-specific biases complicate scRNA-seq data interpretation.
• Existing normalization methods struggle with latent cell type variations and missing data.

Purpose of the Study:

• To introduce an iterative normalization and clustering method for single-cell gene expression data.
• To improve the accuracy of cell type characterization by distinguishing technical noise from biological signals.
• To provide a robust framework for analyzing complex single-cell datasets.

Main Methods:

• Developed a hierarchical Bayesian mixture model with cell-specific scaling factors.
• Implemented an iterative normalization and clustering algorithm.
• Designed a scalable Gibbs inference algorithm for efficient computation.

Main Results:

• The proposed method outperforms global normalization followed by clustering.
• Demonstrated identifiability and weak convergence guarantees for the model.
• Achieved improved cluster inference on both synthetic and real scRNA-seq data.
• Successfully recovered underlying biological structures and cell types.

Conclusions:

• The iterative normalization and clustering method offers superior performance for scRNA-seq data analysis.
• This approach effectively disentangles technical variation from biological signals.
• The method facilitates accurate cell type discovery and characterization.