Skip to content

Machine Learning Guided Material Selection Speeds Discovery



The periodic table is a big menu of elements and their combinations to explore, and manually searching for new materials to decompose combustion byproducts is an endless task.  Members of the Toyota Research Institute of North America’s Materials Research Department (MRD) have worked to develop a Bayesian optimization program to explore a vast material composition space to find new candidates in the quest for a NO decomposition catalyst.  With a small number of initial compounds from which to train, the program was able to guide and focus the efforts of the synthetic team towards unexplored compositions with increased activity.  This combined machine learning- synthetic approach yielded both new directions in chemistry as well as refinement on elemental ratios.  Furthermore, the model itself is not tied specifically to catalysis.  This model can be utilized not only to explore  large compositional spaces of materials for applications where the functionality is driven by chemistry, but for non-materials related optimization problems as well.


A method for more quickly exploring a large portion of the periodic table was developed by blending machine learning with hands-on material synthesis & evaluation



The team is excited to share their work with the greater scientific community by publishing in ACS Catalysis.  Find it at