For proper PGML implementation in materials science, many atomic and ionic properties are used as features. Although many are tabulated, and information is not complete. Furthermore, updates and corrections are needed with time. Using PGML and statistical scrutiny, we try to expand the data by proper interpolation and to enhance accuracy through outlier detection.

In crystals, all atomic interactions are Coulombic. Accordingly, if the geometry and electronic density are known for a given crystal, the electronic structure and all other properties can be calculated by atomic-scale calculations or modeled by physics-guided ML. We aim to develop PGML to predict materials properties solely from chemical composition.