Argonne Team’s ChemGraph Unlocks AI for Chemistry and Materials Science

LEMONT, Ill.--(BUSINESS WIRE)--Computers have made it easier than ever to design materials for specific challenges. Scientists can create virtual versions of materials and simulate how they will behave. But building these atomically precise simulations often requires deep expertise in computational chemistry and complex scientific software.

At the U.S. Department of Energy’s (DOE) Argonne National Laboratory, researchers have developed ChemGraph, an open-source framework that uses artificial intelligence (AI) to help automate computational chemistry workflows. The framework could make advanced simulations more accessible to researchers and students working on challenges in chemistry and materials science, including more efficient combustion, critical materials and next-generation batteries. The team’s work was described recently in the journal Communications Chemistry.

ChemGraph was developed using resources at the Argonne Leadership Computing Facility (ALCF), including the Aurora exascale supercomputer and the ALCF Inference Service, a platform that gives researchers cloud-like access to large language models (LLMs) on the facility’s high-performance computing systems. The ALCF is a DOE Office of Science user facility.

ChemGraph uses LLMs to provide a natural language interface to agent-based automation. A researcher can describe a scientific problem in plain language, and the framework maps that request onto a sequence of computational tasks, software tools and analyses needed to produce a result.

Running these simulations can involve dozens of steps, from choosing the right scientific methods and software to preparing input files, running calculations, analyzing results and refining parameters. ChemGraph assigns different parts of the workflow to AI agents, which act like assistants specializing in tasks such as planning, execution and data aggregation.

Argonne researchers designed ChemGraph to call appropriate scientific tools and libraries, helping reduce the risk of hallucination. Rather than relying on an LLM’s existing knowledge, ChemGraph uses AI to help run physics-based simulations and generate new data.

The team used Aurora to run computationally demanding quantum chemistry simulations, while the ALCF Inference Service provided access to powerful open-weight models on Argonne systems, helping reduce costs and address data-security concerns.

Because ChemGraph is open source, the framework can be adapted to new tasks beyond its initial release. In recent collaborations, Argonne researchers have extended ChemGraph to support spectroscopy simulation and analysis, as well as high-throughput materials screening workflows on Aurora.

ChemGraph also complements DOE’s Genesis Mission, a national initiative to accelerate science through AI. The long-term goal is to make the framework more autonomous, enabling it to plan, execute and refine complex computational workflows with minimal user intervention.