Aug 10 2020
Chemists from RIKEN have developed a machine-learning algorithm that can estimate the compositions of trend-defying new materials. The new algorithm could help find materials for applications in which a trade-off exists between two or more preferred properties.
Machine learning can predict the properties of new materials, thereby saving scientists a lot of time synthesizing them in the laboratory. RIKEN researchers have developed an algorithm that can discover special materials that go against the trend set by other materials. Image Credit: © Monty Rakusen/Getty.
Artificial intelligence holds immense potential to enable researchers to find new materials with useful properties. A machine-learning algorithm trained with the properties and compositions of well-known materials could estimate the properties of unexplored materials, saving a lot of time in the laboratory.
However, identifying new materials for applications can be difficult since there is always a trade-off between two or more material properties. According to Kei Terayama, who was earlier at the RIKEN Center for Advanced Intelligence Project and is now affiliated to Yokohama City University, one such example is the organic materials used for organic solar cells, where it is preferred to optimize both the current and voltage.
There’s a trade-off between voltage and current: a material that exhibits a high voltage will have a low current, whereas one with a high current will have a low voltage.
Kei Terayama, Center for Advanced Intelligence Project, RIKEN
Therefore, material scientists often intend to identify “out-of-trend” materials that buck the usual trade-off. However, traditional machine-learning algorithms are much better at identifying trends than identifying materials that defy those trends.
Terayama and his collaborators have designed a machine-learning algorithm named BLOX (BoundLess Objective free eXploration), which can identify out-of-trend materials.
The research group illustrated the power of the algorithm by employing it to determine eight out-of-trend molecules that exhibit higher photoactivity from a drug-discovery database. The properties of such molecules were in good agreement with those estimated by the algorithm.
We had concerns about the accuracy of the calculation but were delighted to see that the calculation was correct. This shows the potential of computation-driven materials development.
Kei Terayama, Center for Advanced Intelligence Project, RIKEN
BLOX makes use of machine learning to produce a prediction model for crucial material properties. It performs this by integrating data for materials chosen randomly from a materials database with calculation or experimental outcomes.
BLOX employs the model to estimate the properties of a new set of materials. From such new materials, BLOX determines the material that is most different from the entire distribution.
The properties of that material are identified either by calculations or through experiment and then employed to update the machine learning model. The cycle is repeated again.
Most significantly, in contrast to various algorithms developed earlier, BLOX does not impose any limitations on the range of material structures and compositions that can be investigated. Thus, it can reach far and wide in its search for trend-defying materials.
Journal Reference
Terayama, K., et al. (2020) Pushing property limits in materials discovery via boundless objective-free exploration. Chemical Science. doi.org/10.1039/D0SC00982B.