Regularized Optimal Transport for Single-Cell Temporal Trajectory Analysis
Keywords: single-cell transcriptomics, temporal trajectory analysis, optimal transport
TL;DR: We propose a novel framework for temporal trajectory analysis in single-cell transcriptomics problems with regularized optimal transport.
Abstract: The temporal relationship between different cellular states and lineages is only partially understood and has major significance for cell differentiation and cancer progression. However, two pain points persist and limit learning-based solutions: ($a$) lack of real datasets and standardized benchmark for early cell developments; ($b$) the complicated transcriptional data fail classic temporal analyses. We integrate $\texttt{Mouse-RGC}$, a large-scale mouse retinal ganglion cell dataset with annotations for $9$ time stages and $30,000$ gene expressions. Existing approaches show a limited generalization of our datasets. To tackle the modeling bottleneck, we then translate this fundamental biology problem into a machine learning formulation, $\textit{i.e.}$, $\textit{temporal trajectory analysis}$. An innovative regularized optimal transport algorithm, $\texttt{TAROT}$, is proposed to fill in the research gap, consisting of ($1$) customized masked autoencoder to extract high-quality cell representations; ($2$) cost function regularization through biology priors for distribution transports; ($3$) continuous temporal trajectory optimization based on discrete matched time stages. Extensive empirical investigations demonstrate that our framework produces superior cell lineages and pseudotime, compared to existing approaches on $\texttt{Mouse-RGC}$ and another two public benchmarks. Moreover, $\texttt{TAROT}$ is capable of identifying biologically meaningful gene sets along with the developmental trajectory, and its simulated gene knockout results echo the findings in physical wet lab validation.
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: 9423
Loading