Inferring brain plasticity rule under long-term stimulation with structured recurrent dynamics

TL;DR

The Stimulus-Evoked Evolution Recurrent dynamics (STEER) framework is introduced, a dual-timescale model that disentangles fast neural activity from slow plastic changes and provides a data-driven foundation for both mechanistic insight and principled optimization of brain stimulation.

Abstract

Understanding how long-term stimulation reshapes neural circuits requires uncovering the rules of brain plasticity. While short-term synaptic modifications have been extensively characterized, the principles that drive circuit-level reorganization across hours to weeks remain unknown. Here, we formalize these principles as a latent dynamical law that governs how recurrent connectivity evolves under repeated interventions. To capture this law, we introduce the Stimulus-Evoked Evolution Recurrent dynamics (STEER) framework, a dual-timescale model that disentangles fast neural activity from slow plastic changes. STEER represents plasticity as low-dimensional latent coefficients evolving under a learnable recurrence, enabling testable inference of plasticity rules rather than absorbing them into black-box parameters. We validate STEER with four benchmarks: synthetic Lorenz systems with controlled parameter shifts, BCM-based networks with biologically grounded plasticity, a task learning setting with adaptively optimized external stimulation and longitudinal recordings from Parkinsonian rats receiving closed-loop DBS. Our results demonstrate that STEER recovers interpretable update equations, predicts network adaptation under unseen stimulation schedules, and supports the design of improved intervention protocols. By elevating long-term plasticity from a hidden confound to an identifiable dynamical object, STEER provides a data-driven foundation for both mechanistic insight and principled optimization of brain stimulation.

Figures (18)

• Figure 1: Motivation. Repeated stimulation induces slow evolution of network plasticity, with the underlying plasticity rules remaining unknown. The challenge of distinguishing slow plasticity from fast neural dynamics remains underdeveloped, with dynamical systems theory offering a potential framework to model these processes: $\frac{dW}{dt}=f_{\theta}(W)$.
• Figure 1: Stimulus selectivity in simulated neuronal populations. Simulated neuronal populations show clear stimulus selectivity, with neurons responding preferentially to specific stimuli and exhibiting structured co-activation patterns.
• Figure 2: STEER Framework. A dual-timescale, identifiable formulation: within-session dynamics (fast) are generated by structured recurrent connectivity; across-session evolution (slow) follows a stimulus-conditioned latent law that we infer as the plasticity rule.
• Figure 2: Recurrent connection inference. (a) Recurrent weight matrices $\mathbf{W}$ at three time points (sessions 1, 500, and 1000) for the ground truth, STEER, full-rank MD-SSM, and low-rank MD-SSM. (b) Corresponding changes in connectivity between sessions ($\mathbf{\Delta W}$).
• Figure 3: STEER uncovers the parameter evolution in Lorenz system. (a) Plasticity rules for system parameter evolution. (b) Rank selection for a low-rank model, with rank-3 providing optimal approximation. (c) Predicted implicit scaling factor and its similarity to the true plasticity rule (DSA = 0.169). (d) Similarity analysis among STEER and other baseline models (MD-SSM and hierarchical-PLRNN), with STEER having the lowest DSA and best EV (“w.” and “w.o.” denote with and without forecasting, respectively). (e) Prediction results for unseen data (systems 61 and 100) with high explained variance (EV = 0.961 and EV = 0.974).