Understanding the Capabilities of Molecular Graph Neural Networks in Materials Science Through Multimodal Learning and Physical Context Encoding

2025-05-17

AI Summary

Existing molecular graph neural networks (GNNs) predominantly rely on XYZ atomic coordinates, overlooking rich textual chemical information—such as IUPAC names, molecular formulas, and physicochemical properties—abundant in databases like PubChem. To address this, we propose a physics-aware multimodal GNN framework that jointly encodes molecular graphs and textual descriptions. Our method integrates BERT-like text embeddings with explicit physical property encodings and introduces a gated attention mechanism to synergistically model geometric structure and chemical semantics. This work is the first to systematically leverage PubChem’s textual metadata to enhance GNN representations; it further reveals a common limitation of mainstream GNNs: their implicit learning of physical representations lacks robustness and task adaptability. On multiple benchmark datasets, our model achieves significant improvements in predicting electronic properties—including band gaps and ionization energies—with gains exhibiting task-specific patterns.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Molecular graph neural networks (GNNs) often focus exclusively on XYZ-based geometric representations and thus overlook valuable chemical context available in public databases like PubChem. This work introduces a multimodal framework that integrates textual descriptors, such as IUPAC names, molecular formulas, physicochemical properties, and synonyms, alongside molecular graphs. A gated fusion mechanism balances geometric and textual features, allowing models to exploit complementary information. Experiments on benchmark datasets indicate that adding textual data yields notable improvements for certain electronic properties, while gains remain limited for others. Furthermore, the GNN architectures display similar performance patterns (improving and deteriorating on analogous targets), suggesting they learn comparable representations rather than distinctly different physical insights.

Problem

• Molecular GNNs overlook chemical context in databases
• Multimodal framework integrates textual and geometric data
• Textual data improves some electronic property predictions

Innovation

• Multimodal learning integrates textual and geometric data
• Gated fusion balances molecular and textual features
• Textual descriptors enhance electronic property predictions