NJmat: Data-Driven\nMachine Learning Interface to Accelerate\nMaterial Design

Machine\nlearning techniques have significantly transformed the\nway materials scientists conduct research. However, the widespread\ndeployment of machine learning software in daily experimental and\nsimulation research for materials and chemical design has been limited.\nThis is partly due to the substantial time investment and learning\ncurve associated with mastering the necessary codes and computational\nenvironments. In this paper, we introduce a user-friendly, data-driven\nmachine learning interface featuring multiple “button-clicking”\nfunctionalities to streamline the design of materials and chemicals.\nThis interface automates the processes of transforming materials and\nmolecules, performing feature selection, constructing machine learning\nmodels, making virtual predictions, and visualizing results. Such\nautomation accelerates materials prediction and analysis in the inverse\ndesign process, aligning with the time criteria outlined by the Materials\nGenome Initiative. With simple button clicks, researchers can build\nmachine learning models and predict new materials once they have gathered\nexperimental or simulation data. Beyond the ease of use, NJmat offers\nthree additional features for data-driven materials design: (1) automatic\nfeature generation for both inorganic materials (from chemical formulas)\nand organic molecules (from SMILES), (2) automatic generation of Shapley\nplots, and (3) automatic construction of “white-box”\ngenetic models and decision trees to provide scientific insights.\nWe present case studies on surface design for halide perovskite materials\nencompassing both inorganic and organic species. These case studies\nillustrate general machine learning models for virtual predictions\nas well as the automatic featurization and Shapley/genetic model construction\ncapabilities. We anticipate that this software tool will expedite\nmaterials and molecular design within the scope of the Materials Genome\nInitiative, particularly benefiting experimentalists.