Synergizing Physically Constrained MCMC and Chemical-Informed Gaussian Processes for Reaction Network Discovery

2026-06-22

AI Summary

Inferring interpretable reaction networks from sparse, noisy chemical time series is highly challenging due to the strong coupling between discrete network topology and continuous reaction dynamics. This work proposes a gray-box workflow, PC-MCMC-CIGP, which samples plausible reaction topologies via Markov chain Monte Carlo (MCMC) with a spike-and-slab prior, explicitly embedding hard physical constraints such as mass conservation and thermodynamic feasibility. Residual dynamics are modeled using a chemically informed Gaussian process (CIGP), enabling uncertainty-aware network inference and active experimental design. The key innovation lies not in introducing a new model, but in integrating physical constraints into a synergistic MCMC–GP framework. Applied to the H₂+Br₂ system, the method accurately recovers radical pathways; in styrene epoxidation, CIGP-guided Bayesian optimization improves yield by 12.5% over a standard GP-BO baseline, with multi-seed experiments revealing trade-offs among acquisition functions.

Abstract

Extracting interpretable governing equations from sparse, noisy chemical time-series data remains difficult because discrete reaction topology and continuous kinetic parameters are tightly coupled. We present PC-MCMC-CIGP, a reproducible gray-box workflow that combines spike-and-slab topology sampling, hard conservation and thermodynamic screening, and a Chemical-Informed Gaussian Process (CIGP) residual model for parameter calibration and experimental design. The methodological contribution is not a new MCMC or GP family in isolation; rather, it is the integration of these components into a physically constrained workflow with explicit uncertainty-aware acquisition choices. On the H2 + Br2 benchmark, the constrained sampler distinguishes elementary radical pathways from deceptive phenomenological fits in our experiments. On styrene epoxidation, the CIGP optimization loop improves final yield by 12.5% over the reported GP-BO baseline. A new 10-seed acquisition study shows that EI, GWU, PC-EI, uncertainty sampling, discrepancy hunting, and random search have different trade-offs: PC-EI substantially reduces low-yield BO suggestions, while EI-style criteria give the strongest final-yield performance.

Problem

• reaction network discovery
• sparse noisy data
• kinetic parameter identification
• discrete-continuous coupling
• interpretable governing equations

Innovation

• Physically Constrained MCMC
• Chemical-Informed Gaussian Process
• Reaction Network Discovery
• Uncertainty-Aware Acquisition
• Gray-Box Modeling