Group Publishes Work using Graph Attention Networks for Green Chemical Design
We’re thrilled to announce that our recent work “Designing green chemicals by predicting vaporization properties using explainable graph attention networks” was accepted into Green Chemistry.
Here, we show how our lab has developed a new computational approach using a Graph Attention Network (GAT) model to predict key physical properties for renewable energy applications. These properties include heat of vaporization (HoV), critical temperature, flash point, boiling point, and liquid heat capacity. By training the HoV model on a large NIST-WTT database with around 150,000 data points, we achieved high prediction accuracy and reliable uncertainty estimation. Transfer learning techniques applied to smaller datasets for other vaporization properties further reduced errors. The model also provided chemical insights through attention weight analysis and molecular structure interpretation. This methodology, which incorporates uncertainty quantification, transfer learning, and interpretability, is a powerful tool for designing environmentally friendly chemicals for sustainable energy.
To learn more about this work, click here.
