Recently, graph neural networks, applied to some challenging benchmark cases, consistently outperformed state of the art deep neural network methods and other classical machine learning methods.
We are interested in applying graph neural network as a common and universal framework for predicting materials properties starting from molecules. Our group has been active in developing computational methods for predicting transfer integrals and reorganization energies in organic semiconductors. The universality of the method would be tested by transferring it to predict other electronic properties of organic molecules like, HOMO, LUMO energies, band gaps etc. We plan to apply it to the problem of drug discovery as well.