New Workflow Discovers Reaction Networks Using MCMC and Chemical-Informed GPs.

Extracting accurate governing equations from limited and noisy chemical time-series data is a significant challenge, largely due to the intertwined nature of discrete reaction topology and continuous kinetic parameters. Researchers have developed PC-MCMC-CIGP, a novel gray-box workflow designed to address this complexity. This method integrates spike-and-slab topology sampling, rigorous conservation and thermodynamic screening, and a Chemical-Informed Gaussian Process (CIGP) residual model. The core contribution lies not in isolated MCMC or GP families, but in their synergistic integration into a physically constrained workflow that incorporates explicit uncertainty-aware acquisition choices. Evaluations on benchmarks like H2 + Br2 demonstrated the constrained sampler's ability to differentiate elementary radical pathways from misleading phenomenological fits. In styrene epoxidation, the CIGP optimization loop led to a 12.5% improvement in final yield compared to a baseline. Further studies on acquisition strategies revealed that physically constrained criteria significantly reduce low-yield suggestions, while EI-style criteria still offer strong final-yield performance.