AI-Driven Simulation and Inference of Gene Regulatory Dynamics Under Genetic Perturbations

Abstract

Gene regulatory networks (GRNs) encode the causal mechanisms governing dynamic gene expression programs in living systems. Genetic perturbationssuch as knockouts, overexpression, timing shifts, and combinatorial interventionsprovide a powerful means to interrogate these networks by revealing how regulatory influences propagate through time. However, existing approaches for GRN analysis typically separate simulation of regulatory dynamics from inference of network structure, limiting their ability to extract causal information from perturbation-driven time-series data. This work reviews and frames AI-driven methodologies for the joint simulation and inference of GRN dynamics under genetic perturbations. We examine continuous-time and state-space formulations of GRNs, highlighting the challenges posed by high dimensionality, nonlinear dynamics, partial observability, and heterogeneous perturbation regimes. Emerging machine-learning paradigmsincluding deep learning, neural ordinary differential equations, reinforcement learning, graphical models, and counterfactual inferenceare discussed as unifying tools to integrate simulation with causal structure discovery. By leveraging perturbation responses and temporal expression trajectories, these approaches enable partial yet biologically meaningful reconstruction of GRN topology and regulatory strengths. AI-driven GRN modeling under perturbations thus offers a scalable path toward causal understanding of gene regulation, with implications for systems biology, synthetic biology, and precision genetic engineering