Data-driven discovery of methane hydrate promoters

Methane hydrates require extreme conditions, and promoter discovery remains largely empirical. We develop a multimodal deep-learning framework that predicts methane-hydrate equilibrium pressures from molecular structure. Trained on over eighty promoters, the model extrapolates beyond its domain and prospectively identifies ethylene sulfite as a new thermodynamic promoter, experimentally validated within 1 MPa accuracy while forming structure II hydrates.