More and more researchers are looking to artificial intelligence (AI) techniques to accelerate chemical and material development. Venture capital funding and research activity is picking up at a rapid pace, and partnerships and pilot projects are being pursued across the globe. Beyond these top-level trends, however, what is actually receiving R&D attention? Which materials specifically are finding traction within materials informatics research, and who is leading the way?
Using publicly available data, we applied topic modeling algorithms to analyze what is hot in materials informatics and found that the data-driven, quantified results generated by topic modeling agree well with what our human-led research has indicated.
First, we analyzed a variety of data sources to mine and extract insights from, including patents and academic papers. Unfortunately, there are still very few patents to date within materials informatics. Instead, academic papers lead the way in terms of volume, with researchers authoring more than 1,000 materials informatics papers to date, most of them just in the past few years.
To identify where research within materials informatics is clustering, we took two approaches. First, we applied latent Dirichlet allocation (LDA) topic modeling. This machine learning technique processes documents of unstructured text to automatically identify and cluster concepts. Second, we also used a simpler term frequency analysis within the dataset, due to the relatively small number of academic papers within materials informatics.
The results, shown below, highlight that alloys, polymers, and steels are the top three materials classes that attract materials informatics research – at least so far. Rounding out the top five, we see two more specialized and emerging materials classes, namely perovskites and thermoelectrics.
Finally, we also researched who is doing this work by looking within each of the most popular materials informatics clusters to find key players. Here, we focused on the number of research publications authored that touched upon each concept and found an internationally diverse set of innovators. Some institutions and universities, such as the National Institute for Materials Science (NIMS), Caltech, and the University of Cambridge, feature as leaders in many materials informatics categories, while others are more focused on particular material classes.
Companies should take note of the areas where materials informatics is already establishing a strong bench of potential academic partners and form collaborations in the areas relevant to their strategic roadmap.
How does this topic modeling work compared with our own subject matter experts' opinions? Prior to this work, we had already been actively tracking the subareas of materials informatics and have seen similar trends. Alloys, for example, are a group of materials that is relatively well-understood, as we have a long history of metallurgy and relevant data. Plus, the hierarchical structure of alloys is relatively more straightforward and easier to capture experimentally. Therefore, it has become a low-hanging fruit for researchers to test materials informatics on.
- Tech Page: Materials Informatics (for Lux members only)