Predicting thermoelectric properties from chemical formula with explicitly identifying dopant effects

Abstract Dopants play an important role in synthesizing materials to improve target properties or stabilize the materials. In particular, dopants are essential thermoelectic performances of However, existing machine learning methods cannot accurately predict doped due severely nonlinear relations with their properties. Here, we propose a unified architecture neural networks, called DopNet, DopNet identifies effects by explicitly and independently embedding host dopants. our evaluations, outperformed predicting experimentally measured thermoelectric properties, error figure merit (ZT) was 0.06 mean absolute error. significantly effective extrapolation problem that predicts ZTs unknown materials, which is key task discover novel