Transferable Learning of Reaction Pathways from Geometric Priors

2025-04-21

AI Summary

In chemical reaction mechanism studies, conventional minimum energy path (MEP) prediction relies on transition-state geometries or pre-specified reaction coordinates—a major limitation for automated reaction discovery. This work proposes an end-to-end path modeling framework that requires only reactant and product structures, eliminating the need for transition-state annotations. Methodologically, it introduces geodesic interpolation as a geometric prior within an energy-driven training paradigm; employs a symmetry-breaking equivariant neural network to generate variable-length intermediate structures and explicitly regress reaction coordinate deviations; and enforces physical consistency via an energy-based loss function. Evaluated on small-molecule isomerization and [3+2] cycloaddition reactions, the method accurately recovers intrinsic reaction coordinates, achieves substantial gains in path prediction efficiency, and demonstrates strong generalization across diverse reaction types. This approach establishes a new paradigm for scalable exploration of vast reaction spaces.

Technology Category

• Machine Learning: Learning with Manifolds
• Search and Optimization: Learning to Search
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Identifying minimum-energy paths (MEPs) is crucial for understanding chemical reaction mechanisms but remains computationally demanding. We introduce MEPIN, a scalable machine-learning method for efficiently predicting MEPs from reactant and product configurations, without relying on transition-state geometries or pre-optimized reaction paths during training. The task is defined as predicting deviations from geometric interpolations along reaction coordinates. We address this task with a continuous reaction path model based on a symmetry-broken equivariant neural network that generates a flexible number of intermediate structures. The model is trained using an energy-based objective, with efficiency enhanced by incorporating geometric priors from geodesic interpolation as initial interpolations or pre-training objectives. Our approach generalizes across diverse chemical reactions and achieves accurate alignment with reference intrinsic reaction coordinates, as demonstrated on various small molecule reactions and [3+2] cycloadditions. Our method enables the exploration of large chemical reaction spaces with efficient, data-driven predictions of reaction pathways.

Problem

• Predicting minimum-energy paths for chemical reactions efficiently
• Learning reaction pathways without transition-state geometries
• Generalizing across diverse chemical reactions accurately

Innovation

• Machine-learning predicts reaction paths without transition states
• Equivariant neural network generates flexible intermediate structures
• Geometric priors enhance efficiency in path prediction