Saturation-Based Atom Provenance Tracing in Chemical Reaction Networks
By: Marcel Friedrichs, Daniel Merkle
Potential Business Impact:
Tracks atoms through body reactions precisely.
Atom tracing is essential for understanding the fate of labeled atoms in biochemical reaction networks, yet existing computational methods either simplify label correlations or suffer from combinatorial explosion. We introduce a saturation-based framework for enumerating labeling patterns that directly operates on atom-atom maps without requiring flux data or experimental measurements. The approach models reaction semantics using Kleisli morphisms in the powerset monad, allowing for compositional propagation of atom provenance through reaction networks. By iteratively saturating all possible educt combinations of reaction rules, the method exhaustively enumerates labeled molecular configurations, including multiplicities and reuse. Allowing arbitrary initial labeling patterns - including identical or distinct labels - the method expands only isotopomers reachable from these inputs, keeping the configuration space as small as necessary and avoids the full combinatorial growth characteristic of previous approaches. In principle, even every atom could carry a distinct identifier (e.g., tracing all carbon atoms individually), illustrating the generality of the framework beyond practical experimental limitations. The resulting template instance hypergraph captures the complete flow of atoms between compounds and supports projections tailored to experimental targets. Customizable labeling sets significantly reduce generated network sizes, providing efficient and exact atom traces focused on specific compounds or available isotopes. Applications to the tricarboxylic acid cycle, and glycolytic pathways demonstrate that the method fully automatically reproduces known labeling patterns and discovers steady-state labeling behavior. The framework offers a scalable, mechanistically transparent, and generalizable foundation for isotopomer modeling and experiment design.
Similar Papers
Tracking large chemical reaction networks and rare events by neural networks
Molecular Networks
Makes complex science simulations run much faster.
CarAT: Carbon Atom Tracing across Industrial Chemical Value Chains via Chemistry Language Models
Computational Engineering, Finance, and Science
Tracks plant-based carbon in factory products automatically.
Teaching Language Models Mechanistic Explainability Through Arrow-Pushing
Machine Learning (CS)
Teaches computers to predict how chemicals change.