This repository contains the research code for my Master’s thesis on neural surrogate modeling for constraint-based metabolic networks. The project focuses on transformer-based models that predict metabolic fluxes and growth from medium composition inputs by approximating the behavior of flux-based simulations. The aim is to accelerate flux and growth prediction while preserving biological plausibility, and to benchmark predictive accuracy and uncertainty against mechanistic models and experimental datasets.