Mechanistic models of microbial growth aim to explain how cells allocate limited cellular resources to maximize fitness. We developed Growth Balance Analysis (GBA): a mathematical framework for such models, based on (non-linear) enzyme kinetics, mass conservation, and a limited dry mass density. This framework reveals general principles of bacterial physiology, including the marginal costs of cellular components, optimal solutions that oscillate even in constant environments, and the growth-rate dependence of individual components of the translation machinery.

              Motivated by the need to obtain kinetic parameters for such models, we developed a portfolio of general AI models that predict not only turnover numbers (kcat) and Michaelis constants (Km), but also the substrates of enzymes and transporters. These predictions can be improved by multimodal transformer models – analogous to those used to generate images from text prompts – and by models that encode protein 3D structures as molecular images. These predictions can not only help to parameterize theoretical models, but can also provide important guidance for basic research and protein engineering.