Code and data for 'Improved vapor pressure predictions using group contribution-assisted graph convolutional neural networks (GC2NN)'

We propose a novel approach to predict saturation vapor pressures using group contribution-assisted graph convolutional neural networks (GC2NN), which use both, molecular descriptors like molar mass and functional group counts, as well as molecular graphs containing atom and bond features, as representations of molecular structure. Molecular graphs allow the ML model to better infer molecular connectivity and spatial relations compared to methods using other, non-structural embeddings.

We achieve best results with an adaptive-depth GC2NN, where the number of evaluated graph layers depends on molecular size. We apply the model to compounds relevant for the formation of SOA, achieving strong agreement between predicted and experimentally-determined vapor pressure. In this study, we present two models: a general model with broader scope, achieving a mean absolute error (MAE) of 0.69 log-units (R2 = 0.86), and a specialized model focused on atmospheric compounds (MAE = 0.37 log-units, R2 = 0.94).