Prime Implicant Explanations for Reaction Feasibility Prediction

TL;DR

The paper tackles the opacity of RF predictions in computer-aided synthesis planning by introducing formally grounded prime implicant (PI) explanations tailored to reaction graphs encoded as Imaginary Transition State (ITS) graphs. It defines PI reaction explanations as minimally sufficient rooted connected subgraphs containing the reaction center, and leverages an extension DAG to systematically enumerate and prune subgraphs during explanation search. Experiments with a Graph Isomorphism Network on USPTO-derived data show that PI explanations tend to capture core mechanistic factors but often include extra nonessential elements, highlighting both interpretability potential and limitations. The work discusses computational intractability, lack of benchmarks, and directions for improving explanatory quality and efficiency for practical CASP applications.

Abstract

Machine learning models that predict the feasibility of chemical reactions have become central to automated synthesis planning. Despite their predictive success, these models often lack transparency and interpretability. We introduce a novel formulation of prime implicant explanations--also known as minimally sufficient reasons--tailored to this domain, and propose an algorithm for computing such explanations in small-scale reaction prediction tasks. Preliminary experiments demonstrate that our notion of prime implicant explanations conservatively captures the ground truth explanations. That is, such explanations often contain redundant bonds and atoms but consistently capture the molecular attributes that are essential for predicting reaction feasibility.

Figures (3)

• Figure 1: The proposed explanation method provides a minimally sufficient subgraph for the decisions made by a black-box reaction feasibility classifier. This subgraph explanation reflects the chemical mechanism in which $\mathrm{NH}_3$ attacks the ester's electrophilic carbonyl carbon, displacing the alkoxide and forming the more resonance-stabilized amide.
• Figure 2: Two ITS graph visualizations for the reaction in the first row. We'll mostly use the more concise molecular representation on the left but the ITS graph is actually a node and edge labeled graph as shown on the right. The reaction center is the subgraph of changing bonds and is visualized using dotted and dashed lines. Dotted lines indicate breaking bonds, while dashed lines represent newly formed bonds.
• Figure 3: Example of two ITS graphs with isomorphic reaction center. The ITS graph representing the first reaction is feasible whereas the second ITS graph represents an infeasible reaction. The reaction center alone is insufficient to decide the feasibility of a reaction. Some context is needed.

Theorems & Definitions (5)

• definition thmcounterdefinition: Reaction Center
• definition thmcounterdefinition: Subgraph PI explanation Azzolin2025
• remark thmcounterremark
• definition thmcounterdefinition: PI reaction explanation
• definition thmcounterdefinition: Extension DAG