Amiri, M. (2026). Physics-informed deep learning for molecular solubility prediction: integrating thermodynamic constraints with neural network architectures. Scientific Reports. https://doi.org/10.1038/s41598-026-49635-4