Physics‐informed Transfer Learning for Out‐of‐sample Vapor Pressure Predictions

Abstract

Recent advances have enabled machine learning methodologies developed for large datasets to be applied to the small experimental datasets typically available for chemical systems. Such advances typically involve a databased approach to transfer learning, where a portion of the experimental data for the property of interest is used to fine-tune a model that is pre-trained on computationally generated data. This transfer learning approach does not work for very small experimental datasets, where there are only enough data for model validation. Here, we develop a physics-informed transfer learning strategy to train a directed-message passing neural network (D-MPNN) model, enabling extrapolation outside of the training domain. We demonstrate this approach by training a D-MPNN model on interpolated vapor pressures and validate the model on an out-of-sample test set of energetic molecule vapor pressures, achieving accuracy comparable to those of experiments.

Gibbs–Duhem-informed neural networks for binary activity coefficient prediction

“…Several ML methods such as graph neural networks (GNNs), matrix completion methods (MCMs), and transformers have shown great potential for predicting a wide variety of thermophysical properties with high accuracy. This includes both pure component and mixture properties such as solvation free energies, 1 liquid densities 2 and viscosities, 3 vapor pressures, 2,4 solubilities, 5 and fuel ignition indicators 6 A particular focus has recently been placed on using ML for predicting activity coefficients of mixtures due to their high relevance for chemical separation processes. Here, activity coefficients at infinite dilution, 7–9 varying temperature, 10–15 and varying compositions, 16–18 while considering a wide spectrum of molecules, have been targeted with ML, consistently outperforming well-established models such as UNIFAC 19 and COSMO-RS.…”