Integrating and mapping single-cell transcriptomics across the entire gene expression space.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell transcriptomics data is growing exponentially, necessitating integration for large-scale atlases.
• Batch effects introduce biases, hindering accurate data integration and masking biological signals.
• Current methods often focus on highly variable genes, potentially missing information in lowly variable genes.

Purpose of the Study:

• To introduce scGES, a novel deep learning framework for effective batch effect correction in single-cell transcriptomics.
• To address limitations of existing methods by utilizing information from both highly and lowly variable genes.
• To improve the construction of single-cell reference atlases and mapping of query datasets.

Main Methods:

• Developed scGES, a deep learning framework with two models: scGESI for integration and scGESM for mapping.
• Leveraged information from the entire gene expression space, including highly and lowly variable genes.
• Applied comprehensive analyses to real-world single-cell transcriptomics datasets.

Main Results:

• scGES demonstrated superior performance in batch effect correction compared to state-of-the-art methods.
• The framework effectively conserved biological variations within the datasets.
• Utilizing all genes, including lowly variable ones, enhanced downstream analyses and biological insights.

Conclusions:

• scGES offers a robust solution for batch effect correction in single-cell transcriptomics.
• The framework's ability to integrate information from all genes provides broader biological insights.
• scGES advances the construction and utilization of large-scale single-cell reference atlases.