Summary
Measuring how individual cells change over time under different conditions is hard because scientists usually get snapshots of different groups of cells rather than following the same cells. The authors present TempoBridge, a method that uses information from the starting cells to model how cell populations shift with time and treatments more accurately. Compared to existing methods, TempoBridge predicts changes in cell gene activity and overall population patterns better without needing data from the target condition during prediction. This approach helps understand biological changes in cells more precisely by incorporating source cell information and optimal transport techniques.
What this means in practice
- •For biotech data teams: Predict cellular responses to genetic or chemical changes more accurately from available single-cell snapshots without needing target data during inference.
- •For pharmaceutical data scientists: Model drug effects on cell populations using source-conditioned transport models to improve prediction of compound impacts on genes.
Authors
Bowen Han, Lingbei Meng, Shihuan Luo, Yupeng Zang, Wenlin LI, Peize He, Yaodi Luo, Lian Zhang, Jianqing Zhu, Jinchao Xu
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 $L_2$ 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.