Abstract:The use of machine learning methods for accelerating the design of crystalline materials usually requires manually constructed feature vectors or complex transformation of atom coordinates to input the crystal structure, which either constrains the model to certain crystal types or makes it difficult to provide chemical insights. Here, we develop a crystal graph convolutional neural networks framework to directly learn material properties from the connection of atoms in the crystal, providing a universal and interpretable representation of crystalline materials. Our method provides a highly accurate prediction of density functional theory calculated properties for eight different properties of crystals with various structure types and compositions after being trained with $10^4$ data points. Further, our framework is interpretable because one can extract the contributions from local chemical environments to global properties. Using an example of perovskites, we show how this information can be utilized to discover empirical rules for materials design.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to accurately and interpretably predict material properties when accelerating the design of crystalline materials. Specifically, traditional methods usually require manually constructing feature vectors or performing complex transformations on atomic coordinates for inputting crystal structures, which either restricts the model to be applicable only to specific types of crystals or makes it difficult to provide chemical insights. Therefore, the authors developed a framework based on Crystal Graph Convolutional Neural Networks (CGCNN), aiming to directly learn material properties from the connection relationships of atoms in crystals and provide a general and interpretable representation method for crystalline materials. ### Main Problems and Solutions 1. **Problem Description**: - **Manually Constructing Feature Vectors or Complex Transformations**: When dealing with crystalline materials, traditional machine - learning methods usually need to manually construct fixed - length feature vectors or design symmetric - invariant atomic coordinate transformations. The former requires case - by - case design for different properties, and the latter makes it difficult to interpret the model. - **Limited Applicability**: These methods are either restricted to specific types of crystals or difficult to be extended to other types. 2. **Solution**: - **CGCNN Framework**: By constructing a crystal graph, in which nodes represent atoms and edges represent the connection relationships between atoms, features are directly extracted from the crystal structure. - **Automatically Learning Features**: Use a Convolutional Neural Network (CNN) to automatically learn the optimal feature representation from the crystal graph without the need for manual feature design. - **Interpretability**: Provide chemical insights by analyzing the contribution of the local chemical environment to the global properties, making the model interpretable. ### Specific Implementation - **Crystal Graph Representation**: Convert the crystal structure into a graph structure, where nodes represent atoms and edges represent the connections between atoms. Each node and edge is represented by a feature vector, encoding information about atoms and bonds respectively. - **Convolutional Layers and Pooling Layers**: Build a Convolutional Neural Network on the crystal graph, including convolutional layers and pooling layers, which are used to iteratively update the atomic feature vectors and generate the feature vector of the entire crystal. - **Fully - Connected Layers**: Add two fully - connected hidden layers to capture the complex mapping relationship between the crystal structure and properties. - **Output Layer**: Finally, predict the target property through the output layer. ### Advantages - **High Precision**: After training, CGCNN has achieved a precision comparable to DFT calculations in the prediction of 8 different properties. - **Generality**: Applicable to crystalline materials of various structural types and compositions. - **Interpretability**: Able to extract the contribution of the local chemical environment to the global properties, provide chemical insights, and help discover empirical rules for material design. Through these methods, CGCNN not only improves the accuracy of material property prediction but also provides valuable chemical insights, significantly reducing the search space for high - throughput screening.

Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties

Graph Neural Networks for an Accurate and Interpretable Prediction of the Properties of Polycrystalline Materials

CTGNN: Crystal Transformer Graph Neural Network for Crystal Material Property Prediction

Study of crystal properties based on attention mechanism and crystal graph convolutional neural network

General Graph Neural Network-Based Model To Accurately Predict Cocrystal Density and Insight from Data Quality and Feature Representation

Orbital Graph Convolutional Neural Network for Material Property Prediction

Graph Neural Network for Predicting the Effective Properties of Polycrystalline Materials: A Comprehensive Analysis

Coarse-Grained Crystal Graph Neural Networks for Reticular Materials Design

A machine learning‐based crystal graph network and its application in development of functional materials

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

Developing an improved Crystal Graph Convolutional Neural Network framework for accelerated materials discovery

CrysToGraph: A Comprehensive Predictive Model for Crystal Materials Properties and the Benchmark

A geometric-information-enhanced crystal graph network for predicting properties of materials

Improving Crystal Property Prediction from a Multiplex Graph Perspective

Transfer learning for materials informatics using crystal graph convolutional neural network

MT-CGCNN: Integrating Crystal Graph Convolutional Neural Network with Multitask Learning for Material Property Prediction

Study of Crystal Property Prediction Based on Dual Attention Mechanism and Transfer Learning

Predicting Material Properties Using a 3D Graph Neural Network with Invariant Local Descriptors

Material Property Prediction Using Graphs Based on Generically Complete Isometry Invariants

GPT-assisted learning of structure-property relationships by graph neural networks: Application to rare-earth doped phosphors