SCEMENT: scalable and memory efficient integration of large-scale single-cell RNA-sequencing data

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Large-scale single-cell RNA sequencing (scRNA-seq) data integration is crucial for understanding complex biological systems.
• Existing algorithms struggle with scalability for millions of cells and large datasets, often sacrificing accuracy for efficiency.
• Current methods use shortcuts like subsampling or reference datasets, hindering quantitative gene expression analysis.

Purpose of the Study:

• To develop a scalable and memory-efficient algorithm for accurate integration of large-scale scRNA-seq data.
• To overcome the limitations of existing methods in terms of computational cost and accuracy.
• To enable robust downstream analyses requiring precise gene expression information.

Main Methods:

• Introduced SCEMENT (SCalablE and Memory-Efficient iNTegration), a parallel algorithm extending a linear regression model.
• Utilized an unsupervised sparse matrix setting for efficient data integration.
• Implemented the method in C++ for high performance on Linux systems.

Main Results:

• SCEMENT demonstrated superior performance in runtime (up to 214x faster) and memory usage (up to 17.5x less) compared to ComBat, FastIntegration, and Scanorama.
• Successfully integrated millions of cells from tens to hundreds of scRNA-seq datasets in under 25 minutes.
• Facilitated the discovery of rare cell types and improved reconstruction of gene regulatory networks with full quantitative gene expression.

Conclusions:

• SCEMENT provides an accurate, scalable, and memory-efficient solution for large-scale scRNA-seq data integration.
• The method preserves quantitative gene expression information, crucial for in-depth biological analysis.
• SCEMENT enables more robust discovery of biological insights from complex single-cell datasets.