Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ADA-GNN: Atom-Distance-Angle Graph Neural Network for Crystal Material Property Prediction

Jiao Huang, Qianli Xing, Jinglong Ji, Bo Yang

TL;DR

This work tackles crystal-property prediction by introducing ADA-GNN, which explicitly incorporates bond angle information via a dual-scale neighbor partitioning mechanism. By decoupling atom-feature embedding from structure-embedding, ADA-GNN achieves improved training stability and leverages angle information without prohibitive input size, enabling efficient large-scale screening. On Materials Project and JARVIS benchmarks, ADA-GNN delivers state-of-the-art MAE improvements (2.04%-21.82% over PotNet and larger gains over ALiGNN) while reducing inference time relative to angle-aware baselines. The combination of dual-scale modeling and separate embeddings offers a practical approach to leverage angular information for crystal materials with substantial speedups in downstream discovery tasks.

Abstract

Property prediction is a fundamental task in crystal material research. To model atoms and structures, structures represented as graphs are widely used and graph learning-based methods have achieved significant progress. Bond angles and bond distances are two key structural information that greatly influence crystal properties. However, most of the existing works only consider bond distances and overlook bond angles. The main challenge lies in the time cost of handling bond angles, which leads to a significant increase in inference time. To solve this issue, we first propose a crystal structure modeling based on dual scale neighbor partitioning mechanism, which uses a larger scale cutoff for edge neighbors and a smaller scale cutoff for angle neighbors. Then, we propose a novel Atom-Distance-Angle Graph Neural Network (ADA-GNN) for property prediction tasks, which can process node information and structural information separately. The accuracy of predictions and inference time are improved with the dual scale modeling and the specially designed architecture of ADA-GNN. The experimental results validate that our approach achieves state-of-the-art results in two large-scale material benchmark datasets on property prediction tasks.

ADA-GNN: Atom-Distance-Angle Graph Neural Network for Crystal Material Property Prediction

TL;DR

This work tackles crystal-property prediction by introducing ADA-GNN, which explicitly incorporates bond angle information via a dual-scale neighbor partitioning mechanism. By decoupling atom-feature embedding from structure-embedding, ADA-GNN achieves improved training stability and leverages angle information without prohibitive input size, enabling efficient large-scale screening. On Materials Project and JARVIS benchmarks, ADA-GNN delivers state-of-the-art MAE improvements (2.04%-21.82% over PotNet and larger gains over ALiGNN) while reducing inference time relative to angle-aware baselines. The combination of dual-scale modeling and separate embeddings offers a practical approach to leverage angular information for crystal materials with substantial speedups in downstream discovery tasks.

Abstract

Property prediction is a fundamental task in crystal material research. To model atoms and structures, structures represented as graphs are widely used and graph learning-based methods have achieved significant progress. Bond angles and bond distances are two key structural information that greatly influence crystal properties. However, most of the existing works only consider bond distances and overlook bond angles. The main challenge lies in the time cost of handling bond angles, which leads to a significant increase in inference time. To solve this issue, we first propose a crystal structure modeling based on dual scale neighbor partitioning mechanism, which uses a larger scale cutoff for edge neighbors and a smaller scale cutoff for angle neighbors. Then, we propose a novel Atom-Distance-Angle Graph Neural Network (ADA-GNN) for property prediction tasks, which can process node information and structural information separately. The accuracy of predictions and inference time are improved with the dual scale modeling and the specially designed architecture of ADA-GNN. The experimental results validate that our approach achieves state-of-the-art results in two large-scale material benchmark datasets on property prediction tasks.
Paper Structure (20 sections, 15 equations, 3 figures, 5 tables, 1 algorithm)

This paper contains 20 sections, 15 equations, 3 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Dual Scale Modeling of Crystal Materials
  4. Crystal Materials
  5. Dual Scale Neighbor Partitioning Mechanism
  6. Crystal Material Property Prediction
  7. ADA-GNN
  8. Embedding Module
  9. Node Embedding Block
  10. Structure Embedding Block
  11. Message Passing Module
  12. Interaction Block
  13. Readout Module
  14. Training Algorithm and Time Cost Analysis
  15. Experiments
  16. ...and 5 more sections

Figures (3)

  • Figure 1: Illustrations of the crystal structure of NaCl and the bond distances and bond angles. (a): The structural diagram of NaCl. The green sphere signifies the atom Cl, while the purple sphere represents the atom Na. The central cube serves as a visual representation of the unit cell. (b): Bonds in NaCl. (c): Bond distances of bonds. (d): Bond angles formed by the bonds.
  • Figure 2: Illustrations of the partitioning mechanism for edge neighbors and angle neighbors.
  • Figure 3: Illustrations of detailed Architecture of ADA-GNN. (a): The overall structure of ADA-GNN. (b): An illustration of the node embedding module. (c): An illustration of the structure embedding module. (d): An illustration of the interaction block.