Noise performance of stoichiometry-programmed rate-independent CRNs

Determine the performance and robustness to noise of the rate-independent chemical reaction network programming approach that sets neural network parameters via stoichiometric structure (as in Vasić et al. 2022) when implemented in practice; specifically, ascertain how the method behaves under realistic noise in chemical systems (e.g., fluctuations in reaction conditions and species concentrations) when the stoichiometric structures must be updated by chemical species to enable training.

Background

The paper discusses a rate-independent CRN approach where neural network weights are determined solely by stoichiometric structure, independent of reaction rates. Extending this approach to training would require updating feed-forward stoichiometric structures via chemical species, which presents engineering challenges.

The authors explicitly note uncertainty regarding how this stoichiometry-based, rate-independent method would perform under realistic noise, motivating a clear assessment of its robustness in noisy biochemical environments.