EOSnet: Embedded Overlap Structures for Graph Neural Networks in Predicting Material Properties

TL;DR

EOSnet introduces a novel GNN architecture that incorporates Gaussian Overlap Matrix fingerprints as node features, effectively capturing many-body interactions and improving accuracy in predicting various material properties.

Contribution

The paper presents EOSnet, a new GNN model that embeds orbital overlap matrices to better encode many-body atomic interactions without manual feature engineering.

Findings

Achieves a mean absolute error of 0.163 eV in band gap prediction.

97.7% accuracy in metal/non-metal classification.

Outperforms previous models in property prediction tasks.

Abstract

Graph Neural Networks (GNNs) have emerged as powerful tools for predicting material properties, yet they often struggle to capture many-body interactions and require extensive manual feature engineering. Here, we present EOSnet (Embedded Overlap Structures for Graph Neural Networks), a novel approach that addresses these limitations by incorporating Gaussian Overlap Matrix (GOM) fingerprints as node features within the GNN architecture. Unlike models that rely on explicit angular terms or human-engineered features, EOSnet efficiently encodes many-body interactions through orbital overlap matrices, providing a rotationally invariant and transferable representation of atomic environments. The model demonstrates superior performance across various materials property prediction tasks, achieving particularly notable results in properties sensitive to many-body interactions. For band gap prediction, EOSnet achieves a mean absolute error of 0.163 eV, surpassing previous state-of-the-art models. The model also excels in predicting mechanical properties and classifying materials, with 97.7\% accuracy in metal/non-metal classification. These results demonstrate that embedding GOM fingerprints into node features enhances the ability of GNNs to capture complex atomic interactions, making EOSnet a powerful tool for materials discovery and property prediction.