The prediction of boiling points of organosilane compounds is of significant commercial interest in the materials and chemical industry due to their use in semiconductor manufacturing. In this study, we propose the use of directed message-passing neural networks (D-MPNNs) to develop a machine learning model for the prediction of boiling points of organosilane compounds. D-MPNNs have shown promising results in the field of molecular property prediction due to their ability to learn the complex relationships between individual atoms and their neighbors.

We used a dataset of 1700 organosilane compounds with known boiling points to train and evaluate our model. Our dataset was compiled, prepared, and cleaned from publicly available data obtained using the Chemical Abstracts Service. The dataset was split into 80% training, 10% validation, and 10% testing sets. We implemented a D-MPNN architecture that utilized a message passing algorithm to update the features of each atom based on its neighboring atoms, using both prespecified and learned molecular embeddings.

Our model achieved an R² value of 0.98 and a root-mean-squared error (RMSE) of 10.0°C on our test set. Comparison with other state-of-the-art models demonstrated the superior performance of our model. We also compared our results with internal R&D data on company proprietary compounds.

Overall, our study demonstrated the effectiveness of D-MPNNs in predicting the boiling points of organosilane compounds. This approach has the potential to be extended to other molecular property prediction tasks and may provide insights into the underlying relationships between atoms and their bulk properties.

Christopher Cha

Appearances