Integrated Modeling, Observation, and Control of Vitamin B12 Riboswitches: A Hybrid Framework Combining Sliding-Mode Observers, Robust Control, and Machine Learning

Riboswitches are metabolite-sensing RNA elements that regulate gene expression with high specificity, making them central regulators in both natural and synthetic biology. The vitamin B 12 riboswitch in Escherichia coli, which controls the btuB transporter gene, is a canonical model. Despite extensive biochemical insights, robust frameworks for dynamic modeling, state estimation, and control of riboswitch networks remain limited. This work presents an integrated methodology combining deterministic and stochastic modeling, observer design, robust control, and machine learning. We formulate deterministic and stochastic models of btuB regulation under vitamin B 12 availability, benchmark Luenberger observers against Sliding-Mode Observers (SMOs), and show that SMOs converge faster and remain robust under noise and parameter uncertainty. Sliding-Mode Control (SMC) is applied to enforce regulation, maintaining expression trajectories despite disturbances. In parallel, feedforward neural networks provide parameter estimation and soft sensing of hidden molecular states, complementing analytical observers. Finally, the framework is extended to synthetic riboswitches with tunable ligand sensitivity, outlining principles for adaptive genetic circuit design. The proposed framework bridges control theory, systems biology, and machine learning into a unified pipeline. Compared with prior approaches, it achieves higher estimation accuracy, greater robustness to uncertainty, and interpretable integration of machine learning with physicsbased models. This work advances riboswitch research from descriptive characterization toward predictive, tunable, and clinically relevant applications in biosensing, metabolic engineering, and RNA-based therapeutics.