Research map / Probabilistic and scientific modeling / Molecules and proteins
Drift-React: One-step Generation of Reaction Pathways via SE(3) Drifting Fields
Rémi Schlama, Philippe Schwaller
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Drift-React, a one-step generative model for reaction pathways and transition states built on SE(3)-equivariant drifting fields, rather than slow iterative diffusion-style inference.
Problem
Computing minimum energy pathways and transition states with electronic-structure methods needs thousands of costly force evaluations, and existing generative TS models require slow iterative inference and predict only isolated saddle points rather than full pathways.
Approach
Generates entire reaction pathways in one step by learning SE(3)-equivariant drifting fields, instead of iterative denoising, so that the output covers the full MEP and not just a single saddle point.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Drift-React instead of geodesic interpolation | Reaction pathway generation · Halo8 | TS-RMSD (A): 0.316 -> 0.107 | Improved |
| use Drift-React instead of retrained NeuralNEB | Reaction pathway generation · Halo8 | TS-RMSD (A): 0.538 -> 0.107 | Improved |
| use Drift-React instead of geodesic interpolation | Reaction pathway generation · Halo8 | Activation barrier MAE (eV): 1.247 -> 0.914 | Improved |
| use Drift-React instead of linear interpolation | Reaction pathway generation · Transition1x | Per-image energy MAE (eV): ~0.92 -> 3.214 | Hurt |
| use Drift-React instead of linear interpolation | Reaction pathway generation · Transition1x | Activation barrier MAE (eV): 1.881 -> 1.631 | Improved |
| train full pathway (F=10) instead of TS-only (F=3) (ablation) | Drift-React · Transition1x TS-RMSD | TS-RMSD (A): 0.168 -> 0.151 | Improved |
| use Drift-React TS-only instead of ReactOT | Drift-React F=3 · Transition1x TS-only | TS-RMSD (A): 0.103 -> 0.168 | Hurt |
| initialize TS optimization from Drift-React pathway instead of linear midpoint | Drift-React · Halo8 TS refinement | Median optimization cycles: 16 -> 10 | Improved |
| initialize TS optimization from Drift-React pathway instead of linear midpoint | Drift-React · Halo8 TS refinement | Mean absolute barrier error (eV): 0.143 -> 0.501 | Hurt |
Closest papers by meaning
- Action-Minimization Meets Generative Modeling: Efficient Transition Path Sampling with the Onsager-Machlup Functional (ICML 2025 · similarity 0.89)
- Learning Condensed Graph via Differentiable Atom Mapping for Reaction Yield Prediction (ICML 2025 · similarity 0.87)
- RETRO SYNFLOW: Discrete Flow-Matching for Accurate and Diverse Single-Step Retrosynthesis (NeurIPS 2025 · similarity 0.87)
- FlashMD: long-stride, universal prediction of molecular dynamics (NeurIPS 2025 · similarity 0.86)
- Generative Modeling of Molecular Dynamics Trajectories (NeurIPS 2024 · similarity 0.86)
- Transition Path Sampling with Improved Off-Policy Training of Diffusion Path Samplers (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
