FreeBridge: Variational Schrödinger Bridges for Cellular Transition Dynamics

2026-06-09

AI Summary

This work proposes a Schrödinger bridge–based approach to reconstruct biologically meaningful continuous single-cell trajectories under the constraint of observing only the marginal distributions of cell populations before and after perturbation. By defining atomic states as single-cell representations obtained via instance segmentation, the method constructs a fixed cell manifold and introduces an empirical latent-space support regularization mechanism that confines the stochastic transport process strictly within the manifold spanned by observed cellular morphologies, thereby ensuring biological interpretability of intermediate states. Integrating variational Schrödinger bridges with stochastic optimal transport, the approach maintains or improves endpoint alignment accuracy and mechanism retention on the BBBC021, RxRx1, and JUMP datasets, while significantly reducing violations of the cellular morphology support in intermediate states on BBBC021.

Abstract

High-content imaging assays quantify cellular responses to chemical and genetic perturbations, yet continuous trajectories of individual cells are unobservable because cells are chemically fixed at acquisition. Perturbation modeling therefore reduces to inferring stochastic transport between control and treated populations observed only as separate marginals. While recent generative models achieve strong end-point alignment, boundary consistency does not determine intermediate evolution: multiple stochastic processes may connect identical marginals while traversing regions unsupported by observed single-cell morphologies. We introduce \textbf{FreeBridge}, a Schrödinger Bridge formulation for single-cell transition modeling under endpoint-only supervision. FreeBridge defines atomic states as instance-segmented single-cell representations, establishing a fixed cellular manifold, and learns stochastic transport constrained within this geometry via empirical latent support regularization. Across BBBC021, RxRx1, and JUMP, FreeBridge maintains competitive or improved endpoint fidelity and mechanism-of-action retention under a unified evaluation protocol; on BBBC021, it further reduces intermediate support violations. These findings highlight the importance of geometric grounding for biologically interpretable perturbation dynamics. Project page: https://y-research-sbu.github.io/FreeBridge/.

Problem

• cellular transition dynamics
• stochastic transport
• single-cell imaging
• endpoint-only supervision
• morphological support

Innovation

• Schrödinger Bridge
• single-cell dynamics
• geometric regularization
• stochastic transport
• latent support