Preserving Composition and Crystal Structures of Chemical Compounds in Atomic Embedding

We develop a new representation learning method in the chemistry domain. Given a large set of compounds of inorganic crystals, the extraction model learns the embeddings of atoms so that the predictive model can place them into the periodic table correctly. Our method preserves not only the compounds' compositions but also their crystal structures. Experiments demonstrate the effectiveness of the proposed method, compared to the state-of-the-art method (in PNAS 2018).

I. Introduction

In this work, we investigate the important role of structure information in describing the properties of the chemical compounds. The top part of Figure 1 shows two types of information of chemical compounds, composition (on the left) and crystal structures (on the right). Different chemical compounds may have the same composition but totally different crystal structures. For example, the Wurtzite and Sphalerite are of the same composition (ZnS”) but different crystal systems: Wurtzite is hexagonal and Sphalerite is cubic. We focus on solving two problems: (1) preserving both the composition and crystal structure information of chemical compounds in the embeddings of atoms, (2) investigating which granularity of the crystal structure generates the most effective embeddings.