Teaching Language Models Mechanistic Explainability Through Arrow-Pushing

2025-12-05

AI Summary

Existing computer-aided synthesis planning (CASP) systems lack mechanistic foundations, hindering reliable assessment of synthetic feasibility. Method: We propose MechSMILES—a formal, electron-flow–based encoding of reaction mechanisms—and introduce the first physically constrained, interpretable mechanism prediction framework grounded in conservation laws. Integrating language models with MechSMILES, we systematically train on mech-USPTO-31k and FlowER across four progressive mechanistic prediction tasks: step prediction, template extraction, atom mapping, and full-mechanism retrieval. Results: Our approach achieves >95% Top-3 accuracy in elementary step prediction, >73% overall score on mech-USPTO-31k, and 93% full-mechanism retrieval accuracy on FlowER. This work establishes the first architecture-agnostic mechanistic prediction benchmark, enabling catalyst-aware template extraction and reaction validity verification—thereby substantially enhancing the chemical reasonability and interpretability of CASP.

Technology Category

• Multiagent Systems: Mechanism Design
• Cognitive Modeling & Cognitive Systems: Computational Creativity
• Machine Learning: Neuro-Symbolic Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Chemical reaction mechanisms provide crucial insight into synthesizability, yet current Computer-Assisted Synthesis Planning (CASP) systems lack mechanistic grounding. We introduce a computational framework for teaching language models to predict chemical reaction mechanisms through arrow pushing formalism, a century-old notation that tracks electron flow while respecting conservation laws. We developed MechSMILES, a compact textual format encoding molecular structure and electron flow, and trained language models on four mechanism prediction tasks of increasing complexity using mechanistic reaction datasets, such as mech-USPTO-31k and FlowER. Our models achieve more than 95% top-3 accuracy on elementary step prediction and scores that surpass 73% on mech-USPTO-31k, and 93% on FlowER dataset for the retrieval of complete reaction mechanisms on our hardest task. This mechanistic understanding enables three key applications. First, our models serve as post-hoc validators for CASP systems, filtering chemically implausible transformations. Second, they enable holistic atom-to-atom mapping that tracks all atoms, including hydrogens. Third, they extract catalyst-aware reaction templates that distinguish recycled catalysts from spectator species. By grounding predictions in physically meaningful electron moves that ensure conservation of mass and charge, this work provides a pathway toward more explainable and chemically valid computational synthesis planning, while providing an architecture-agnostic framework for the benchmarking of mechanism prediction.

Problem

• Teaching language models to predict chemical reaction mechanisms using arrow-pushing formalism.
• Developing a computational framework for explainable and chemically valid synthesis planning.
• Enabling applications like validating CASP systems and extracting catalyst-aware reaction templates.

Innovation

• Teaching language models arrow-pushing formalism for mechanistic prediction
• Developing MechSMILES format to encode molecular structure and electron flow
• Enabling applications like CASP validation and atom-to-atom mapping