Optimal Transport in Single-Cell Multi-Omics Integration: Advances, Challenges, Limitations, and Future Directions

Introduction: Optimal Transport (OT) theory, with its capability to quantify and minimize distributional discrepancies in a geometry-aware manner, has emerged as a powerful framework for aligning heterogeneous modalities in single-cell multi-omics studies. Recently, OT-based tools have demonstrated significant potential across applications such as cross-modal alignment, dynamic trajectory inference, and spatial multi-omics integration. Methods: This review surveys OT-based methods for single-cell integration and categorizes their applications across tasks such as cross-modal alignment, spatial transcriptomics mapping, and dynamic trajectory inference. We further introduce hybrid strategies incorporating graph neural networks, neural parameterization, and biological priors into OT formulations. Results: In this study, we highlight recent methodological advances—ranging from Gromov- Wasserstein OT to neural OT models—and propose key future directions, including interpretable OT frameworks and deep learning-augmented transport models. Discussion: Despite its advantages, we identify several key challenges related to computational scalability, integration of biological priors, and interpretability that hinder broader adoption, and highlight potential solutions, including neural OT, the use of biological networks, and interpretable factorization of coupling matrices. Conclusion: OT provides a flexible and scalable paradigm for single-cell multi-omics integration, offering unique advantages in handling heterogeneity, noise, and feature mismatch. With continued advances in efficiency, biological interpretability, and clinical relevance, OT is poised to become a foundational approach for integrative single-cell analysis and precision medicine.