Decoding Gene Regulatory Networks from Single-Cell RNA Velocity
arXiv:2608.09722
2026
Dynamics
2 ideas extracted · analyzed Sep 1, 2026
What the math gives to ML
The paper provides a transferable mechanism for recovering sparse dynamical couplings from partial state observations: passive trajectories may be structurally non-identifying, whereas controlled perturbations can make the interaction matrix identifiable. Its integral reconstruction formulation replaces noisy numerical differentiation with regression on time-integrated states, enabling finite-sample sparse-recovery guarantees. The most promising neural-network transfer is a derivative-free sparse identification and training procedure for neural ODEs or recurrent state-space models, combined with active perturbations chosen to maximize an empirical excitation condition.
Ideas from this paper
✗ Mechanism failed
2026
Expose a recurrent model to deliberately designed input pulses or latent-state perturbations instead of training only on passive trajectories. Choose perturbations that maximize the smallest eigenvalue of the accumulated feature Gramian, making otherwise indistinguishable recurrent couplings recoverable and reducing uncertainty in long-horizon predictions.
Useful7/10
Difficulty6/10
Novelty8/10
✗ Failed on benchmark
2026
Train a recurrent or neural-ODE state transition with an integral residual instead of matching noisy finite-difference derivatives. Enforce sparse regulator-to-state connectivity with group sparsity, so the model learns a compact dynamical mechanism while avoiding the severe variance amplification caused by estimating derivatives from sampled data.
Useful7/10
Difficulty5/10
Novelty7/10