Catalyst design is a challenging matter as catalysis involves multiple complex factors including element combinations and responses against experimental conditions. Catalysts are typically designed by experimental researchers through trial-and-error where recent advancements in first principles calculations also provide assistance to the design process. For cases like this, catalyst informatics is introduced where data science techniques such as machine learning are used to design catalysts from catalyst data. While machine learning is a powerful approach, it does not unveil how the machine learns from the data, making it very difficult to understand the reasoning behind why certain predictions are made. Here, catalyst big data is transformed into semantic networks in order to visualize and represent underlying relational information that are useful for catalyst design. In particular, oxidative coupling of methane (OCM) catalysts big data is created using high throughput experiments where the OCM reaction involves the direct conversion of methane to etylane/ethane (C2). By transforming such data into complex networks, it becomes possible to visually understand trends such as tendencies of sets of elements to correlate with specific ranges of C2 yield as well as sets of elements whose reactivity may be affected by the elements they are paired with. These networks are then used to design catalysts with potentially high C2 yields and are experimentally tested via high throughput experiments where the majority of the suggested catalysts result in high C2 yields. This demonstrates that creating networks from catalysts big data can be an alternative way to design catalysts and, more importantly, visualize the underlying details of how the reaction may be occurring with regards to chemical elements and experimental conditions. Thus, an alternative way of designing catalysts from knowledge network is introduced with greater efficiency with a higher possibility of success than typically seen with traditional catalyst design methods.

Creating semantic networks from catalyst big data for designing active catalysts

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.