Unpaired Single-Cell Dataset Alignment with Wavelet Optimal Transport
Keywords: single cell, optimal transport, unpaired dataset alignment, spectral graph wavelets, gromov wasserstein
TL;DR: We propose a new multi-resolution optimal transport method to align highly noisy, incomplete, and non-isometric single-cell datasets.
Abstract: Aligning single-cell samples across different datasets and modalities is an important task with the rise of high-throughput single-cell technologies. Currently, collecting multi-modality datasets with paired samples is difficult, expensive, and impossible in some cases, motivating methods to align unpaired samples from distinct uni-modality datasets. While dataset alignment problems have been addressed in various domains, single-cell data introduce additional complexity including high levels of noise, dropout, and non-isometry between data spaces. In response to these unique challenges, we propose Wavelet Optimal Transport (WOT), a multi-resolution optimal transport method that aligns samples by minimizing the spectral graph wavelet discrepancies across datasets. Filters are incorporated into the optimization process to eliminate non-essential scales and wavelets, enhancing the quality of correspondences. We demonstrate the capacity of WOT in highly noisy and non-isometric conditions, outperforming previous state-of-the-art methods by significant margins, especially on real single-cell datasets.
Supplementary Material: zip
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7769
Loading