PhyFold: Environment Aware Physics Informed Neural Network for Protein Folding Dynamics

Predicting protein folding remains one of biology’s most complex challenges, particularly under varying environmental conditions. This study presents PhyFold, a hybrid modeling framework that integrates a reaction diffusion based formulation with Physics Informed Neural Networks (PINNs) and deep learning embeddings to capture the spatiotemporal dynamics of protein folding. Unlike purely data driven models, PhyFold explicitly encodes physical constraints and environmental factors temperature, pH, and pressure within the learning process. Using fluorescence data as a surrogate for experimental observables, the model achieves dual functionality, predicting folding dynamics and functional states. Mathematical validation against finite difference solutions confirms physical consistency, while biological validation demonstrates strong alignment with experimental fluorescence responses. By bridging data driven inference and physical understanding, PhyFold provides an interpretable and generalizable framework for exploring protein behavior across diverse physiological conditions, advancing integrative modeling at the interface of machine learning and biophysics.