Parallel clustering of single cell transcriptomic data with split-merge sampling on Dirichlet process mixtures.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Massive single-cell transcriptome data is now available due to droplet-based systems.
• This data enables analysis of cellular and molecular processes at single-cell resolution.
• Current clustering methods struggle with accuracy, requiring prior knowledge of cluster numbers, and computational speed.

Purpose of the Study:

• To address the limitations of existing clustering methods for single-cell transcriptome data.
• To develop a more accurate, efficient, and flexible clustering model.
• To enable robust analysis of large-scale single-cell genomic datasets.

Main Methods:

• Development of the Parallelized Split Merge Sampling on Dirichlet Process Mixture Model (Para-DPMM).
• Utilizing a cluster-level sampling mechanism for improved convergence and optimality.
• Leveraging high-performance computing (HPC) clusters for parallelized massive inference.

Main Results:

• The Para-DPMM model demonstrates superior clustering quality compared to existing methods.
• The model achieves significantly faster computational speeds for large datasets.
• Experimental results validate the effectiveness of the split-merge sampling and parallelization.

Conclusions:

• The Para-DPMM model offers a powerful solution for analyzing massive single-cell transcriptome data.
• It overcomes key challenges in clustering quality, cluster number determination, and computational efficiency.
• The model's performance and speed make it suitable for large-scale biological research.