AI and the World's Hard Problems · Life sciences · Nature (machine learning) · 2026-10-21
Deep dynamical models of single-cell multiomic velocities predict loss-of-function and rescue perturbations in B cells
DynaVelo, a generative neural ordinary differential equation model, learns the joint dynamics of gene expression and transcription factor motif activity from single-cell multiome data, allowing perturbation in silico. It recovered the bifurcating in vivo dynamics of wild-type mouse germinal center B cells and revealed how these dynamics change under heterozygous loss of Arid1a and Ctcf. In silico perturbation predicted Ctcf, Stat3, and Bcl6 as rescue targets for Arid1a loss, and in vivo experiments in double-heterozygous mice confirmed rescue in all three cases.