Scalable batch-correction approach for integrating large-scale single-cell transcriptomes.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Integrating large-scale single-cell transcriptomes from diverse sources presents significant computational challenges.
• Existing batch-correction methods often struggle with scalability for massive datasets.

Purpose of the Study:

• To introduce Fugue, a novel and efficient batch-correction method designed for super large-scale single-cell transcriptome integration.
• To demonstrate Fugue's scalability and effectiveness in harmonizing data from heterogeneous sources.

Main Methods:

• Fugue encodes batch information as trainable parameters added to single-cell expression profiles.
• A contrastive learning approach is employed to learn feature representations from these additive profiles.
• The method's scalability is validated by integrating the entire Human Cell Atlas dataset.

Main Results:

• Fugue demonstrates superior scalability, successfully integrating all single cells from the Human Cell Atlas.
• Benchmarking shows Fugue outperforms state-of-the-art methods in data alignment.
• Fugue consistently improves clustering preservation across integrated datasets.

Conclusions:

• Fugue provides a simple, efficient, and scalable solution for large-scale single-cell data integration.
• The proposed method enhances data harmonization and preserves biological structure.
• This work facilitates the integration of ever-growing single-cell transcriptomic datasets.