Break

The production and consumption of fossil fuels contributes significantly to environmental pollution and climate change. A recent effective alternative is to utilize zeolites, nanoporous materials, to act as a sustainable catalyst in innovative separations processes of transforming biomass into chemicals and fuels because it generates considerably less pollution than traditional methods. When synthesizing a suitable zeolite, an OSDA (Organic Structure Directing Agents) and its binding energy with zeolite guide the formation of the zeolite's porous structure. To guide zeolite synthesis experimental work, computational techniques in calculating binding energy between zeolites and OSDAs have been utilized. However, this is an extremely expensive and resource-intensive process, especially in the scope of 6 hundred billion possible pairings between over 3 million OSDAs and 85 thousand zeolites. Therefore, there is a need to build a machine-learning model to predict the binding energy between zeolites and OSDAs to save computational time and resources. Our project aims to develop a streamlined codebase that employs active learning to continuously modify and improve the existing machine learning model. The code will enable the system to automatically screen the chemical space for data points of interest and set up calculations of actual binding energies. Then it will extract these computed energies from the database and retrain the model with those new data. This process iterates until the model is capable of robust inference in the desired chemical space of zeolites and OSDAs. This exploration could lead to the identification of suitable syntheses of new, desired zeolites that have desirable properties for catalyzing these sustainable chemical processes.