LoCoFlow: Local-Context Flow Matching for Single-Cell Dynamics and Fate Inference
Single-cell sequencing profiles cellular states at discrete time points, providing population-level snapshots from which developmental and disease-associated trajectories can be reconstructed. Existing optimal-transport and flow-matching approaches typically infer temporal couplings at the level of individual cells, with limited representation of the local manifold geometry of cell-state space…
LoCoFlow, a flow-matching framework, has been developed to enhance single-cell dynamics and fate inference from sequencing profiles. Unlike traditional optimal-transport and flow-matching methods that infer temporal couplings at the individual cell level, LoCoFlow incorporates local point-cloud geometry and conditional fate uncertainty.
Hierarchical transport between populations is achieved through overlapping patches, which are regularized within the patches using an endpoint-preserving neural bridge. This design supervises conditional fields with patch-shared and cell-specific latent variables, enabling the sampling of alternative trajectories without requiring future observations.
Studies comparing LoCoFlow to scDiffEq on temporal and spatial benchmarks have revealed improvements in transport distance. When applied to lineage-traced LARRY data, LoCoFlow demonstrates a reduction in the mean raw Brier error from 0.7990 to 0.6620 when using the same number of endpoint draws. The framework's application to expression perturbations further highlights the distinct roles of focal-cell and neighborhood information.
Geometric-patch models exhibit aligned background and whole-patch responses, while random-minibatch models show opposing focal and background responses. These findings underscore the utility of local context as a representation for probabilistic fate inference, offering testable hypotheses regarding how it influences predicted responses.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
