Leenhouts, RoelRoelLeenhoutsJankelevitch, SebastienSebastienJankelevitchRaike, RoelRoelRaikeMüller, SimonSimonMüllerVermeire, FlorenceFlorenceVermeire2025-11-052025-11-052025-07Systems and Control Transactions 4: 1662-1669 (2025)978-1-7779403-3-1https://hdl.handle.net/11420/58461Phase transition enthalpies, such as those for fusion, vaporization, and sublimation, are vital for understanding thermodynamic properties and aiding early-stage process design. However, measuring these properties is often time-consuming and costly, leading to increased interest in computational methods for fast and accurate predictions. Graph neural networks (GNNs), known for their ability to learn complex molecular representations, have emerged as state-of-the-art tools for predicting various thermophysical properties. Despite their success, GNNs do not inherently obey thermodynamic laws. In this study, we present a multitask GNN designed to predict vaporization, fusion, and sublimation enthalpies of organic compounds. We modified the loss function of the GNN, accounting for the thermodynamic cycle of the three phase transition enthalpies. To train the model, we digitized the extensive Chickos and Acree compendium, which encompasses 32,023 experimental measurements. Two approaches were explored: soft constraints, which guide the model toward thermodynamic consistency, and hard constraints, which enforce fully consistent predictions. The GNN achieved root mean squared errors (RMSEs) of 19.9 kJ/mol for sublimation, 11.0 kJ/mol for fusion, and 16.5 kJ/mol for vaporization enthalpies on the test set. Soft constraints were found to provide a good balance between accuracy and thermodynamic consistency, whereas hard constraints prioritized fidelity at the expense of predictive performance. When compared to the conventional Joback group contribution method the GNN demonstrated an improved accuracy and applicability range. This work underscores the potential of thermodynamics-informed GNNs for predicting thermodynamic properties accurately while maintaining consistency, paving the way for more reliable and efficient computational approaches.enhttps://creativecommons.org/licenses/by-sa/4.0/Graph neural networksProperty predictionPhysics informedPhase transition enthalpiesComputer Science, Information and General Works::004: Computer SciencesNatural Sciences and Mathematics::541: Physical; TheoreticalJournal Articlehttps://doi.org/10.15480/882.1608610.69997/sct.14063810.15480/882.16086Journal Article