Deep dynamical models of single-cell multiomic velocities predict loss-of-function and rescue perturbations in B cells.

DynaVelo accurately models single-cell gene expression and TF motif dynamics, revealing bifurcating germinal center B cell trajectories and perturbation effects.

• Why it matters: Understanding gene regulatory network changes in B cells under genetic perturbations is crucial for insights into immune function and disease, yet dynamic modeling of these processes is limited.
• What they did: The authors developed DynaVelo, a neural ordinary differential equation model trained on single-cell multiome data, to learn joint gene expression and TF activity dynamics and simulate perturbations.
• The result: DynaVelo recovers known and predicts disrupted TF-gene interactions in mutants, identifying rescue targets like Ctcf, Stat3, and Bcl6, with in vivo validation confirming phenotype rescue and dynamics shift.