GraphAgents: Knowledge Graph-Guided Agentic AI for Cross-Domain Materials Design
ACS Materials Au · 11.11.2026
DOI: 10.1021/acsmaterialsau.6c00100 ↗Kurz gefasst
Fachartikel in der Zeitschrift ACS Materials Au, online erschienen am 11.11.2026. Verfasst von Isabella A. Stewart und 3 weiteren Autor:innen. Die Arbeit nennt 56 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Advanced Scientific Computing Research, Saint-Gobain Northboro Research and Development Center.
Abstract (Originalsprache)
Large language models (LLMs) promise to accelerate discovery by reasoning across the ever-expanding landscape of scientific knowledge. Yet, as the reservoir of knowledge grows, the challenge is no longer access to information but rather the ability to connect it in meaningful, domain-spanning ways. In materials science, where discoveries and new hypotheses emerge from bridging insights across disciplines and demand the integration of concepts from molecular chemistry to mechanical performance, this challenge is especially acute. Neither human researchers nor single-agent LLMs can contend with this torrent of information, with the latter often producing brittle reasoning and hallucinations. To address this bottleneck, we introduce a multiagent framework guided by large-scale knowledge graphs to address the urgent challenge of finding sustainable substitutes for per- and polyfluoroalkyl substances (PFAS), a class of synthetic chemicals widely used in industrial and consumer applications that are the subject of increasing regulatory scrutiny due to concerns raised in scientific literature and evolving policy efforts to ban certain PFAS compounds. Agents in the framework specialize in problem decomposition, evidence retrieval, design parameter extraction, and graph traversal, uncovering latent connections across distinct pockets of scientific knowledge that support downstream hypothesis generation for PFAS-free materials. Our ablation studies show that the full multiagent pipeline outperforms single-shot prompting, underscoring the value of distributed specialization and relational reasoning. We show that by tailoring graph traversal strategies, the system can alternate between exploitative searches that focus hypothesis generation on domain-critical outcomes and exploratory searches that surface emergent cross-connections in the broader network. Shown through the exemplar of biomedical tubing, the framework generates sustainable material alternatives to PFAS-based tubing that balance tribological performance, thermal stability, chemical resistance, and biocompatibility. This work establishes a framework that combines knowledge graphs with multiagent reasoning to expand the materials design space and reveal interdisciplinary connections. We showcase several initial PFAS-free design candidates to demonstrate the approach.
Förderung
- Advanced Scientific Computing Research
- Saint-Gobain Northboro Research and Development Center
Zitieren
APA
Stewart, I. A., Hage, T. P., Hsu, Y., & Buehler, M. J. (2026). GraphAgents: Knowledge Graph-Guided Agentic AI for Cross-Domain Materials Design. ACS Materials Au. https://doi.org/10.1021/acsmaterialsau.6c00100
Vancouver
Stewart IA, Hage TP, Hsu Y, Buehler MJ. GraphAgents: Knowledge Graph-Guided Agentic AI for Cross-Domain Materials Design. ACS Materials Au. 2026. doi:10.1021/acsmaterialsau.6c00100
BibTeX
@article{Stewart2026,
author = {Stewart, Isabella A. and Hage, Tarjei Paule and Hsu, Yu-Chuan and Buehler, Markus J.},
title = {GraphAgents: Knowledge Graph-Guided Agentic AI for Cross-Domain Materials Design},
journal = {ACS Materials Au},
year = {2026},
doi = {10.1021/acsmaterialsau.6c00100}
}
EUVIMED