TempoBridge: Source-Conditioned Flow Matching with Optimal Transport Couplings for Single-Cell Population Transitions
Abstract
Destructive single-cell measurements provide unpaired population snapshots rather than observations of the same cells across conditions. Local cell states and transition requests may also be insufficient to distinguish responses across source populations. We introduce TempoBridge, a common source-conditioned transport formulation for temporal, genetic, and chemical population transitions. Source cells initialize latent transport and provide a fixed empirical population summary. The velocity field receives this summary alongside the evolving cell state, flow time, and a structured transition descriptor. Minibatch optimal transport (OT) supplies couplings only for conditional flow-matching training paths; inference requires neither target expression nor OT computation. On held-out donors, TempoBridge achieves an Energy distance of 0.129 versus 0.144 for scGen. Genetic mean-expression L2 error is 2.261 versus 3.156 for scGPT-scratch under Seen 2/2. On held-out compounds, condition-averaged drug-effect correlation is 0.598 versus 0.561 for the CellFlow adapter. Temporal ablations show higher mean distributional error after removing source context, replacing optimal transport with random pairing, or replacing flow matching with static residual regression. Together, these results demonstrate the predictive utility of a common source-conditioned transport formulation across held-out donors, gene combinations, and compounds.
Related
- Neural-DISCO: Source-Conditioned Counterfactual Editing of Neural Population Activity
- Population fragmentation and the weak Allee effect: critical transitions caused by increased connectivity
- Opposing reward-epoch population coding in anterior insular cortex dissociates sign- and goal-tracking despite equivalent conditioned responding
- Identifiable inverse optimal transport for single-cell state transitions
- A risk-benefit assessment of moderate plant-forward dietary transitions in the Danish population
BibTeX
@misc{han2026tempobridge,
title = {{TempoBridge: Source-Conditioned Flow Matching with Optimal Transport Couplings for Single-Cell Population Transitions}},
author = {Bowen Han and Lingbei Meng and Shihuan Luo and Yupeng Zang and Wenlin LI and Peize He and Yaodi Luo and Lian Zhang and Jianqing Zhu and Jinchao Xu},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.04945},
eprint = {2610.04945},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.04945},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - TempoBridge: Source-Conditioned Flow Matching with Optimal Transport Couplings for Single-Cell Population Transitions
AU - Bowen Han
AU - Lingbei Meng
AU - Shihuan Luo
AU - Yupeng Zang
AU - Wenlin LI
AU - Peize He
AU - Yaodi Luo
AU - Lian Zhang
AU - Jianqing Zhu
AU - Jinchao Xu
PY - 2026
DA - 2026/10/10
PB - arXiv
DO - 10.48550/arxiv.2610.04945
UR - https://arxiv.org/abs/2610.04945
AB - Destructive single-cell measurements provide unpaired population snapshots rather than observations of the same cells across conditions. Local cell states and transition requests may also be insufficient to distinguish responses across source populations. We introduce TempoBridge, a common source-conditioned transport formulation for temporal, genetic, and chemical population transitions. Source cells initialize latent transport and provide a fixed empirical population summary. The velocity field receives this summary alongside the evolving cell state, flow time, and a structured transition descriptor. Minibatch optimal transport (OT) supplies couplings only for conditional flow-matching training paths; inference requires neither target expression nor OT computation. On held-out donors, TempoBridge achieves an Energy distance of 0.129 versus 0.144 for scGen. Genetic mean-expression L_2 error is 2.261 versus 3.156 for scGPT-scratch under Seen 2/2. On held-out compounds, condition-averaged drug-effect correlation is 0.598 versus 0.561 for the CellFlow adapter. Temporal ablations show higher mean distributional error after removing source context, replacing optimal transport with random pairing, or replacing flow matching with static residual regression. Together, these results demonstrate the predictive utility of a common source-conditioned transport formulation across held-out donors, gene combinations, and compounds.
ER -