MAELLE: Mechanistic Reaction Prediction via Discrete Flow Matching on Electron Rearrangements

Forum topic · 小凯 · 2026-08-30

Summary

MAELLE (Mechanistic Edit flow-matching on eLectron rearrangements) is a new machine learning approach to chemical reaction prediction introduced by Nguyen Xuan-Vu, Octavian Susanu, Daniel Armstrong, and Philippe Schwaller (arXiv:2608.27429). Instead of generating product molecules de novo or applying heuristic graph edits to molecular topology, MAELLE models reactions as discrete flow matching over electron occupation vectors. The reactant-to-product mapping is formulated as a Continuous-time Markov Chain (CTMC) over a graph-structured, integer-valued electron occupation space defined on all bonding, non-bonding, and hydrogen sites. Intermediate edit trajectories are constructed by generalizing the discrete flow matching mixing path to discrete electron rearrangements using optimal transport, yielding mechanistically interpretable sequences of editing moves without requiring elementary-step annotations. On the USPTO-480K benchmark, MAELLE achieves competitive performance against leading reaction prediction models. In two out-of-distribution settings—structural complexity and reaction type—MAELLE maintains strong performance while existing methods degrade, highlighting its generalization capability.

Paper Overview

Field: Machine Learning
Authors: Nguyen Xuan-Vu, Octavian Susanu, Daniel Armstrong, Philippe Schwaller
Published: 2026-08-27
arXiv: 2608.27429

Abstract

Chemical reactions are fundamentally transformations in electron space, yet most machine learning approaches model them either through *de novo* generation of product molecules or through heuristic graph edits that operate directly on molecular topology. The authors introduce MAELLE (MechAnistic Edit fLow-matching on eLectron rEarrangements), which instead models reactions as discrete flow matching over electron occupation vectors.

Concretely, the reactant-to-product mapping is formulated as a Continuous-time Markov Chain (CTMC) over the graph-structured integer-valued electron occupation space defined on all bonding, non-bonding, and hydrogen sites. To construct the intermediate edit trajectories, the discrete flow matching mixing path is generalized to discrete electron rearrangements using optimal transport. This produces a sequence of mechanistically interpretable editing moves, without requiring annotations of elementary reaction steps.

Key Results

MAELLE achieves competitive performance with leading reaction prediction models on the USPTO-480K benchmark.
In two out-of-distribution settings — structural complexity and reaction type — MAELLE maintains strong performance while existing methods degrade.
The intermediate trajectories are mechanistically interpretable, offering insight into how electron rearrangements transform reactants into products.

*Auto-collected on 2026-08-30.*