AI Research

Mechanistic Reaction Prediction via Discrete Flow M... | AI Research

Key Takeaways

Mechanistic Reaction Prediction via Discrete Flow Matching on Graph-Structured Electron Occupation explained MAELLE (MechAnistic Edit fLow-matching on eLectr...
MAELLE achieves competitive performance on the USPTO-480K benchmark compared with leading reaction prediction models.
Finally, because the learned flow operates over the full electron redistribution, MAELLE naturally recovers mechanistic trajectories that align with known chemistry and can predict side products of a reaction.
# Mechanistic Reaction Prediction via Discrete Flow Matching on Graph-Structured Electron Occupation explained
By modeling reactions as a sequence of electron rearrangements, the researchers aim to provide a more accurate and interpretable way to predict reaction outcomes and side products.

Paper AbstractExpand

Chemical reactions are fundamentally transformations in electron space, yet most machine learning approaches model them either through \textit{de novo} generation of product molecules or through heuristic graph edits that operate directly on molecular topology. We introduce MAELLE (\textbf{M}ech\textbf{A}nistic \textbf{E}dit f\textbf{L}ow-matching on e\textbf{L}ectron r\textbf{E}arrangements), which instead models reactions as discrete flow matching over electron occupation vectors. Concretely, we formulate the reactant-to-product mapping 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, we generalize the discrete flow matching mixture path to discrete electron rearrangements using Optimal Transport, yielding a sequence of mechanistically interpretable edit moves without requiring elementary step annotations. MAELLE achieves competitive performance on the USPTO-480K benchmark compared with leading reaction prediction models. Beyond in-distribution accuracy, we evaluate robustness across two out-of-distribution settings - structural complexity and reaction type - and find that MAELLE maintains strong performance where existing methods degrade. Finally, because the learned flow operates over the full electron redistribution, MAELLE naturally recovers mechanistic trajectories that align with known chemistry and can predict side products of a reaction.

Mechanistic Reaction Prediction via Discrete Flow Matching on Graph-Structured Electron Occupation explained

MAELLE (MechAnistic Edit fLow-matching on eLectron rEarrangements) is a machine learning model designed to predict chemical reactions by focusing on the movement of electrons rather than just changes in molecular structure. By modeling reactions as a sequence of electron rearrangements, the researchers aim to provide a more accurate and interpretable way to predict reaction outcomes and side products.

Modeling reactions as electron flow

Most existing machine learning models for reaction prediction rely on generating product molecules from scratch or applying heuristic edits to molecular graphs. MAELLE takes a different approach by treating reactions as transformations in electron space. The researchers formulate the mapping from reactants to products as a Continuous-time Markov Chain (CTMC). This model operates on an electron occupation vector, which tracks the number of electrons across all bonding, non-bonding, and hydrogen sites in a molecule.

How MAELLE constructs trajectories

To determine how a reaction proceeds, MAELLE uses discrete flow matching. The researchers generalized this technique to model electron rearrangements using Optimal Transport. This allows the model to generate a sequence of intermediate "edit moves" that describe how electrons shift from the reactant state to the product state. A key advantage of this method is that it does not require researchers to provide annotations for elementary reaction steps; the model learns these trajectories automatically.

Performance and robustness

The researchers evaluated MAELLE on the USPTO-480K benchmark, where it achieved performance competitive with leading reaction prediction models. Beyond standard accuracy, the model was tested for robustness in out-of-distribution settings, specifically regarding reaction types and structural complexity. In these scenarios, MAELLE maintained strong performance where other existing methods showed degradation.

Mechanistic insights

Because MAELLE tracks the full redistribution of electrons, it can recover mechanistic trajectories that align with established chemical principles. This capability allows the model to predict not only the primary product of a reaction but also potential side products. This suggests that by focusing on the underlying electron movement, the model captures a more realistic representation of chemical transformations than methods that only look at the final molecular topology.

Comments (0)

No comments yet

Be the first to share your thoughts!