LAUSR

Abstract A combined experimental and theoretical study of the phase behavior of mixtures of gelatin and maltodextrin is presented. This system, at neutral pH and 50∘C, exhibits coexistence of three equilibrium phases: two liquid phases and a third, which is a solid-like precipitate that resembles a coacervate. This phase behavior completely changes at lower pH, where only two phases remain. Flory-Huggins solution theory is used to model the observed phase behavior. The optimal fits of the associated Flory-Huggins mixing parameters are obtained using a novel non-linear regression method based on neural networks. The neural network was trained using as input only the number of phases present, without knowledge of the specific compositions of the coexisting phases. The optimal Flory-Huggins parameters result in phase diagrams which are consistent with our experimental data. A comparison with the experimental results indicates that the method can produce relevant fits of the model parameters in situations where the information is limited. Moreover, the formation of the solid-like precipitate can be explained within the standard Flory-Huggins framework, despite the fact that important local and specific interactions or induced structural changes occur in the system to produce such a precipitate. In conclusion, the model obtained for this system allows a straightforward exploration, prior to the experimental validation, of the appropriate conditions for future practical applications of this system.