About the Seminar:
Our group’s overarching goal is to develop new methods to extract sustainable fuels, polymers, and chemicals from plants. Our approach has been to develop and apply computational tools to biological and chemical conversion processes as part of an iterative ‘model-validate-predict’ design process for de novo catalysts.
With its high carbon and hydrogen content, lignocellulosic biomass presents an alternative to petroleum as a nearly carbon-neutral precursor to upgraded liquid fuels, renewable polymers, and chemicals. I will show some representative results in designing new catalysts for biological and chemocatalytic processes of biomass using theoretical and machine learning approaches.
Solubility is a crucial chemical property that affects the reactivity in chemical synthesis and the separation ability between solvents and solutes during the extraction and purification of products. Its importance has been considered a key factor when designing chemical processes in various fields, such as electrochemistry, organic, pharmaceutical, petroleum, and sustainable polymer chemistry. Here, we first introduce solubility prediction of organic compounds via self-evolving solubility databases and graph neural networks (GNNs). Prediction of polymer and protein solubility followed by experimental validation are other practical applications. This project resulted in extensive, self-evolving, and reliable solubility databases and accurate GNNs for various solutes, including polymers and proteins dissolved in single and multicomponent solvents.
Then, we will go over our traditional “Fuel property first” design approach to reduce emissions and increase performance for biofuel candidates. We have developed chemistry and physics-informed GNN models to predict a few fuel properties, including yield sooting index (YSI), cetane number (CN), critical temperature (Tc), and heat of vaporization (HoV). These properties are key factors in determining fuel performance, emissions, and safety and can vary substantially between bioblendstock candidates under consideration. The model and methodology used in this work can be applied to other fuel properties, leading to rational principles for designing high-performance fuels.
