Because collecting precise and accurate chemistry data is often challenging, chemistry datasets usually populate only a small region of the chemical space, which limits the performance and the scope of application of a data-driven model.  To address this issue, we integrated an active learning machine with automatic ab initio calculations to realize a self-evolving model that can continuously adapt to the species appointed by the users.  In the present work, we demonstrate the self-evolving concept by modeling the formation enthalpies of polycyclic species.  By combining a molecular graph convolutional neural network with a dropout training strategy, the model we developed can predict enthalpies for a broad range of polycyclic species and estimate uncertainty in each enthalpy value.  For the species which the current model is uncertain about, the automatic ab initio calculations provide accurate enthalpies to improve the performance of the model.  For a test set composed of 2,858 polycyclic (and cyclic) hydrocarbons and oxygenates, the enthalpies predicted by the model agree with the reference values with an RMS error of 2.63 kcal/mol.  We found that a model originally trained on hydrocarbons and oxygenates can broaden its prediction coverage to nitrogen-containing species via the active learning process, suggesting that the continuous learning strategy is not only able to improve the model accuracy but also expand the predictive capacity of a model to an unseen species domain.

Self-evolving machine: A continuously improved model for molecular thermochemistry

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.