CelLink: integrating single-cell multi-omics data with weak feature linkage and imbalanced cell populations.

Summary

CelLink introduces an optimal-transport–based, multi-phase pipeline that integrates single-cell modalities under weak feature linkage and imbalanced populations, excluding unreliable matches to prevent error propagation. Benchmarks show superior mixing, manifold preservation, and feature imputation, uniquely enabling transcriptome imputation for spatial proteomics to support spatial endocrine biology.

Key Findings

• Introduces a multi-phase optimal transport pipeline with normalization/smoothing and dynamic exclusion of unmapped cells to handle weak linkage and imbalanced populations.
• Outperforms state-of-the-art methods on data mixing, manifold preservation, and feature imputation across scRNA-seq, spatial proteomics, and CITE-seq benchmarks.
• Enables transcriptomic profile imputation from spatial proteomics, supporting spatial transcriptomic analyses and correction of mislabeled cells.

Clinical Implications

While methodological, CelLink enables higher-fidelity cellular maps of endocrine organs, improving target discovery, subtype annotation, and spatial context for disease processes (e.g., islet autoimmunity, thyroid cancer microenvironment).

Limitations

• Performance may depend on parameter choices and dataset-specific preprocessing.
• Primarily computational validation; limited wet-lab orthogonal validation of imputed features.

Future Directions

Apply CelLink to endocrine organ atlases (islet, thyroid, pituitary) and integrate with perturbational datasets; develop uncertainty quantification for imputations and prospective experimental validation.