NJmat: Data-Driven Machine Learning Interface to Accelerate Material Design.

Area of Science:

• Materials Science
• Computational Chemistry
• Data Science

Background:

• Machine learning (ML) adoption in materials science is hindered by steep learning curves and extensive coding requirements.
• Current ML tools demand significant time investment, limiting their integration into daily research workflows for materials and chemical design.

Purpose of the Study:

• To introduce NJmat, a novel, user-friendly, data-driven ML interface designed to streamline materials and chemical design.
• To automate key ML processes, including data transformation, feature selection, model construction, prediction, and result visualization.

Main Methods:

• Development of a graphical user interface (GUI) with "button-clicking" functionalities for ML model building and prediction.
• Implementation of automated featurization for inorganic materials (from chemical formulas) and organic molecules (from SMILES strings).
• Integration of automatic generation for Shapley plots and "white-box" genetic models/decision trees for scientific insights.

Main Results:

• NJmat successfully automates the entire ML pipeline for materials and chemical design, significantly reducing the time and expertise needed.
• Case studies on halide perovskite surface design demonstrate the tool's efficacy in virtual prediction, automated featurization, and interpretable model construction.
• The software facilitates rapid inverse design, aligning with the Materials Genome Initiative's goals.

Conclusions:

• NJmat democratizes data-driven materials design, empowering researchers, especially experimentalists, to leverage ML without extensive computational expertise.
• The automated features and "white-box" models provide both speed and scientific understanding, accelerating the discovery of new materials and chemicals.
• This tool is expected to significantly expedite research within the Materials Genome Initiative framework.