AnnFlux: object-conditioned neural stochastic differential equations for single-cell perturbation dynamics
Single-cell perturbation profiling measures responses to genetic and chemical interventions, yet most models learn a static map, ignoring how populations move over time and how perturbations combine. AnnFlux, an object-conditioned stochastic differential equation, learns a drift field in latent cell-state space. Conditioning on the perturbing object makes the field queryable one object at a time,…
Single-cell perturbation profiling measures cellular responses to genetic and chemical interventions. However, most models used for this purpose learn a static map, failing to account for how cell populations evolve over time and how multiple perturbations interact. AnnFlux, a novel approach, learns a drift field in the latent cell-state space.
By conditioning on the specific perturbing object, AnnFlux can query the field one object at a time, resulting in per-object drifts that are comparable across genes and drugs. This tailored drift field allows AnnFlux to interpolate data from a held-out timepoint in an epithelial-mesenchymal transition time course and predict unseen perturbations.
Furthermore, AnnFlux enhances distributional fidelity and accurately predicts responses to combinations of held-out perturbations. An example of its predictive power is demonstrated by a predicted IFN-response signature that was later found to be associated with transcriptional landscape proximity in an independent pan-cancer spatial atlas.
Ultimately, AnnFlux offers a framework for mapping the perturbation-driven evolution of cell states as continuous trajectories, and represents unseen perturbations using prior-knowledge embeddings.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
