Model synthesis and identifiability analysis of stiff chemical reaction systems with inVAErt networks

Researchers from arXiv applied inVAErt networks, a VAE-based invertible framework, to tackle stiff chemical kinetics ODE systems spanning 2-20 equations, 3-20 species, and 3-25 rate parameters. The approach combines conditional ResNet and LSTM emulators with structured inverse inference, recovering reaction rates, integration time, and initial conditions from species concentration snapshots. Relative RMSE metrics demonstrate strong identifiability on both reversible and irreversible kinetics. Computational chemists and systems biologists building surrogate models for kinetic parameter estimation should prioritize this over traditional adjoint-based or MCMC methods for ill-posed inverse problems.