scGALA advances graph link prediction-based cell alignment for comprehensive data integration and harmonization

Guo Jiang ^1,2, Kailu Song ^2,3, Gregory J Fonseca ⁴

¹Department of Medicine, Division of Experimental Medicine, McGill University, Montreal, QC, Canada.
²Meakins-Christie Laboratories, Research Institute of the McGill University Health Centre, McGill University, Montreal, QC, Canada.
³Quantitative Life Sciences, McGill University, Montreal, QC, Canada.
⁴Department of Medical Sciences, Khalifa University, Abu Dhabi, United Arab Emirates.
⁵Lung Bioengineering and Regeneration, Department of Experimental Medical Sciences, Faculty of Medicine, Lund University, Lund, Sweden.
⁶Lund Stem Cell Center, Faculty of Medicine, Lund University, Lund, Sweden.
⁷Department of Medicine and Biomedical Engineering, McGill University, Montreal, QC, Canada.
⁸Department of Bioengineering, College of Engineering, California Institute for Quantitative Biosciences, University of California Berkeley, Berkeley, CA, USA.
⁹Ann Romney Center for Neurologic Diseases, Brigham and Women's Hospital, Harvard Medical School, Boston, MA, USA.
¹⁰State Key Laboratory of Systems Medicine for Cancer, Center for Single-Cell Omics, School of Public Health, Shanghai Jiao Tong University School of Medicine, Shanghai, China. huiwang@shsmu.edu.cn.
¹¹Department of Medicine, Division of Experimental Medicine, McGill University, Montreal, QC, Canada. jun.ding@mcgill.ca.
¹²Meakins-Christie Laboratories, Research Institute of the McGill University Health Centre, McGill University, Montreal, QC, Canada. jun.ding@mcgill.ca.
¹³Quantitative Life Sciences, McGill University, Montreal, QC, Canada. jun.ding@mcgill.ca.
¹⁴School of Computer Science, McGill University, Montreal, QC, Canada. jun.ding@mcgill.ca.
¹⁵Mila - Quebec AI Institute, Montreal, QC, Canada. jun.ding@mcgill.ca.

⁰Department of Medicine, Division of Experimental Medicine, McGill University, Montreal, QC, Canada.

Nature Communications +

|

November 26, 2025

View abstract on PubMed

Summary

scGALA, a new graph-based learning framework, enhances single-cell data integration by improving cell alignment. It achieves more accurate cell correspondences across diverse datasets, boosting downstream analysis tasks.

Area Of Science

Computational biology
Genomics
Bioinformatics

Background

Single-cell technologies enable multimodal data acquisition, revealing cellular heterogeneity.
Robust cell alignment is crucial for integrating and harmonizing single-cell data, including batch correction and multi-omics integration.
Existing alignment methods often rely on rigid metrics, limiting accuracy across diverse cell populations.

Purpose Of The Study

To introduce scGALA, a novel graph-based learning framework for redefining cell alignment in single-cell data integration.
To overcome limitations of existing methods by developing a flexible and accurate cell alignment strategy.
To enhance the performance of downstream single-cell data integration tasks.

Main Methods

scGALA utilizes graph attention networks and a score-driven, task-independent optimization strategy.
It constructs enriched graphs by integrating gene expression with auxiliary data (e.g., spatial coordinates).
Alignment is refined through self-supervised graph link prediction using a deep neural network.

Main Results

scGALA identifies over 25% more high-confidence cell alignments compared to existing methods.
The framework maintains or improves accuracy in cell correspondence.
Demonstrated versatility in enhancing various single-cell data integration tasks.

Conclusions

scGALA offers a significant advancement in cell alignment for single-cell data integration.
The graph-based approach effectively captures complex cell-cell relationships.
Improved alignment via scGALA facilitates more robust and accurate downstream analyses.

Related Concept Videos

Overview of Cell-Matrix Interactions

The extracellular matrix or ECM holds cells together to form a tissue and allows the cells within the tissue to communicate. ECM comprises proteins such as fibronectin, collagen, laminin, etc. The most abundant protein in this space is collagen. Collagen fibers are interwoven with carbohydrate-containing protein molecules called proteoglycans. ECM allows cell migration and provides a structural scaffold at cell adhesion that anchors the cell when the extracellular matrix proteins interact with...