Accurate generation of chemical reaction transition states by conditional flow matching

2025-07-14

AI Summary

Chemical reaction transition states (TSs) are experimentally inaccessible due to their ultrashort lifetimes, and traditional density functional theory (DFT)-based iterative TS optimization is computationally prohibitive for high-throughput reaction network exploration. To address this, we propose TS-GEN, the first model to employ conditional flow matching for TS generation: given reactant and product conformations as conditions and a Gaussian prior as the starting distribution, it learns a deterministic, single-step mapping from noise to TS geometry via optimal transport. TS-GEN bypasses iterative optimization entirely, enabling inference in just 0.06 seconds on GPU. Generated TS structures achieve sub-angstrom accuracy (RMSE = 0.004 Å), with a mean barrier energy error of 1.019 kcal/mol; over 87% of predictions meet chemical accuracy (≤1 kcal/mol). The method thus delivers unprecedented efficiency, accuracy, and generalization for automated TS prediction.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Transition state (TS) structures define the critical geometries and energy barriers underlying chemical reactivity, yet their fleeting nature renders them experimentally elusive and drives the reliance on costly, high-throughput density functional theory (DFT) calculations. Here, we introduce TS-GEN, a conditional flow-matching generative model that maps samples from a simple Gaussian prior directly to transition-state saddle-point geometries in a single, deterministic pass. By embedding both reactant and product conformations as conditioning information, TS-GEN learns to transport latent noise to true TS structures via an optimal-transport path, effectively replacing the iterative optimization common in nudged-elastic band or string-method algorithms. TS-GEN delivers unprecedented accuracy, achieving a root-mean-square deviation of $0.004 m{mathring{A}}$ (vs. $0.103 m{mathring{A}}$ for prior state-of-the-art) and a mean barrier-height error of $1.019 { m kcal/mol}$ (vs. $2.864 { m kcal/mol}$), while requiring only $0.06 { m s}$ GPU time per inference. Over 87% of generated TSs meet chemical-accuracy criteria ($<1.58 { m kcal/mol}$ error), substantially outpacing existing methods. TS-GEN also exhibits strong transferability to out-of-distribution reactions from a larger database. By uniting sub-angstrom precision, sub-second speed, and broad applicability, TS-GEN will be highly useful for high-throughput exploration of complex reaction networks, paving the way to the exploration of novel chemical reaction mechanisms.

Problem

• Accurately predicting transition state structures in chemical reactions
• Reducing reliance on costly DFT calculations for TS analysis
• Improving speed and precision in reaction mechanism exploration

Innovation

• Generative model maps Gaussian prior to transition states
• Embeds reactant and product conformations as conditioning
• Replaces iterative optimization with optimal-transport path