EAGER: ADAPT: Optimizing chemical reaction networks with AI

The proposal focuses on optimizing prototypical “toy model” CRNs for the first two years. These toy models are well-studied playgrounds useful for developing and characterizing the best AI algorithms for CRN optimization. we anticipate that the methods and understanding will rapidly translate from toy models into more complex, detailed models of reaction-diffusion chemistry like the model illustrated in Fig. 1. If the methods we propose are able to scale, a particular impactful application will be to optimize both protein and gene regulatory networks. In both cases, the standard models are essentially CRNs, with or without diffusion. Experimental technique often allow one to alter rates in those CRNs by, for example, overexpressing or underexpressing a single component of the network, and the dynamical consequences of that alteration are nonlinear and hard to extract from the stochastic kinetics.