xChemAgents: Agentic AI for Explainable Quantum Chemistry

2025-05-26

AI Summary

In quantum chemical property prediction, naive concatenation of multimodal descriptors (geometric structures and textual representations) degrades performance on symmetry-sensitive tasks and compromises interpretability. To address this, we propose a physics-aware dual-agent collaborative framework: the Selector agent employs a sparse weighting mechanism to adaptively identify salient chemical descriptors and generate natural-language reasoning; the Validator agent iteratively incorporates physical constraints—such as unit consistency and scaling laws—via dialog-based refinement. The method integrates large language model (LLM)-driven agent collaboration, multimodal graph neural networks, and dynamic descriptor selection. On standard benchmarks, it achieves up to 22% reduction in mean absolute error over strong baselines. Crucially, it produces high-fidelity, physically verifiable chemical explanations—embedding interpretability intrinsically within the prediction pipeline—and simultaneously advances both predictive accuracy and human-understandable reasoning.

Technology Category

• Knowledge Representation and Reasoning: Qualitative Reasoning
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Agentic search
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Recent progress in multimodal graph neural networks has demonstrated that augmenting atomic XYZ geometries with textual chemical descriptors can enhance predictive accuracy across a range of electronic and thermodynamic properties. However, naively appending large sets of heterogeneous descriptors often degrades performance on tasks sensitive to molecular shape or symmetry, and undermines interpretability. xChemAgents proposes a cooperative agent framework that injects physics-aware reasoning into multimodal property prediction. xChemAgents comprises two language-model-based agents: a Selector, which adaptively identifies a sparse, weighted subset of descriptors relevant to each target, and provides a natural language rationale; and a Validator, which enforces physical constraints such as unit consistency and scaling laws through iterative dialogue. On standard benchmark datasets, xChemAgents achieves up to a 22% reduction in mean absolute error over strong baselines, while producing faithful, human-interpretable explanations. Experiment results highlight the potential of cooperative, self-verifying agents to enhance both accuracy and transparency in foundation-model-driven materials science. The implementation and accompanying dataset are available anonymously at https://github.com/KurbanIntelligenceLab/xChemAgents.

Problem

• Enhancing quantum chemistry predictions with multimodal descriptors
• Addressing performance degradation from heterogeneous descriptor usage
• Improving interpretability in physics-aware property prediction

Innovation

• Multimodal graph neural networks with textual descriptors
• Physics-aware cooperative agent framework
• Selector and Validator agents for interpretability