Nanoengineers at the University of California San Diego have developed an AI algorithm that predicts the structure and dynamic properties of any material - whether existing or new, almost instantaneously. Known as M3GNet, the algorithm was used to develop a database of more than 31 million yet-to-besynthesized materials with properties predicted by machine learning algorithms; the database is called matterverse.ai. Matterverse.ai facilitates the discovery of new technological materials with exceptional properties.
As explained recently in the journal Nature Computational Science, the properties of a material are determined by the arrangement of its atoms. However, existing approaches to obtain that arrangement have either been prohibitively expensive or ineffective for many elements. Similar to proteins, we need to know the structure of a material to predict its properties. In short, what we need is an AlphaFold for materials.
AlphaFold is an AI algorithm developed by Google DeepMind to predict protein structure. To build the equivalent for materials, the team combined "mathematical-graph neural networks" with "many-body interactions" to build a deep learning architecture that works universally, with high accuracy, across all the elements of the periodic table. Mathematical graphs” are natural representations of a collection of atoms. Using mathematical graphs, researchers can represent the full complexity of materials without being subject to the combinatorial explosion of terms associated with traditional formalisms.
To train their model, the team used the huge database of material energies, forces and stresses collected by the so-called Materials Project over the past decade. The result is the M3GNet interatomic potential (or IAP), which can predict the energies and forces in any collection of atoms. Matterverse.ai was generated through combinatorial elemental substitutions on more than 5,000 structural prototypes in the Inorganic Crystal Structure Database.
The M3GNet IAP was then used to obtain the equilibrium crystal structure for property prediction. Of the 31 million materials in matterverse.ai today, more than a million are predicted to be potentially stable. The team intends to greatly expand not just the number of materials, but also the number of predicted properties, including high-value properties with small data sizes using a multi-fidelity approach they developed earlier.
The M3GNet IAP also has broad applications in dynamic simulations of materials and property predictions as well. To promote the use of M3GNet, the team has released the framework as open-source Python code on Github. And there are plans to integrate the M3GNet IAP as a tool in commercial materials simulation packages.