Material Property Prediction with Element Attribute Knowledge Graphs and Multimodal Representation Learning

TL;DR

A multimodal fusion framework, ESNet, is proposed, which integrates element property features with crystal structure features to generate joint multimodal representations, enabling the model to consider both microstructural composition and chemical characteristics of the materials.

Abstract

Machine learning has become a crucial tool for predicting the properties of crystalline materials. However, existing methods primarily represent material information by constructing multi-edge graphs of crystal structures, often overlooking the chemical and physical properties of elements (such as atomic radius, electronegativity, melting point, and ionization energy), which have a significant impact on material performance. To address this limitation, we first constructed an element property knowledge graph and utilized an embedding model to encode the element attributes within the knowledge graph. Furthermore, we propose a multimodal fusion framework, ESNet, which integrates element property features with crystal structure features to generate joint multimodal representations. This provides a more comprehensive perspective for predicting the performance of crystalline materials, enabling the model to consider both microstructural composition and chemical characteristics of the materials. We conducted experiments on the Materials Project benchmark dataset, which showed leading performance in the bandgap prediction task and achieved results on a par with existing benchmarks in the formation energy prediction task.

Figures (6)

• Figure 1: Overview of our adopted methodology ESNet.The complete element embedding is obtained by weighting and summing the individual element vectors output by the Element KG Encoder based on the elemental ratios in the crystal structure.The number of modules n in the Fusion Encoder is set according to the specific task.
• Figure 2: Examples of Fe element related attributes
• Figure 3: Overview of ElementKG Construction And Embedding. a. Construction of the Elemental Knowledge Graph. We collect element attribute knowledge from the periodic table and construct element triples. b.Embedding process for elemental knowledge graph. Three documents (structural, lexical and combinatorial) are derived from the elemental knowledge graph and merged into a single document for training the word embedding model Word2Vec. This process integrates the elemental attribute knowledge into a unified representation that facilitates the prediction of downstream crystal structure properties.
• Figure 4: Illustration of the iComFormer pipeline. a. Materialcrystal structure.b. Crystal multimodal graph.c. iComFormer architecture, where elements of the same color belong to the same module.
• Figure 5: "H" atoms vector in CGCNN. The blue dashed boxes in the figure are used to distinguish other elements from the lanthanide and actinide elements. There are the following issues with the encoding of the hydrogen atom "H": a. The electronegativity is 2.2, so the value in the red dashed box should be set to 1. b. The Valence electrons property is not labeled. c. The "Atomic volume" property should belong to 10 categories, but the data is missing the two values marked by the black dashed box.