Finding Pathways in Reaction Networks guided by Energy Barriers using Integer Linear Programming
Adittya Pal, Rolf Fagerberg, Jakob Lykke Andersen, Peter Dittrich, Daniel Merkle · 2025
Read Finding Pathways in Reaction Networks guided by Energy Barriers using Integer Linear Programming by Adittya Pal, Rolf Fagerberg, Jakob Lykke Andersen, Peter Dittrich, Daniel Merkle (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-10609.txt.
Introduction
Adittya Pal and co-authors present a computer method in a 2025 paper for finding routes through chemical reaction networks. Chemists often need a chain of reactions that turns simple starting materials into a target molecule. Each step has an energy barrier, and a good route should favor steps that are likely to happen.
The method uses integer linear programming, a math technique for picking the best option under set rules, applied to networks drawn as branching diagrams. A scoring rule ranks rival routes by how probable they are, based on the physics of the barriers. The team also built a pipeline that estimates barrier heights for networks that lack them. They tested the method on a small network and found routes to glycine and a related acid.
Catalog record
| CREATOR | Adittya Pal, Rolf Fagerberg, Jakob Lykke Andersen, Peter Dittrich, Daniel Merkle |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-10609
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Finding Pathways in Reaction Networks guided by Energy Barriers using Integer Linear Programming Adittya Pal, Rolf Fagerberg, Jakob Lykke Andersen, Peter Dittrich, Daniel Merkle Abstract Analyzing synthesis pathways for target molecules in a chemical reaction network annotated with information on the kinetics of individual reactions is an area of active study. This work presents a computational methodology for searching for pathways in reaction networks which is based on integer linear programming and the modeling of reaction networks by directed hypergraphs. Often multiple pathways fit the given search criteria. To rank them, we develop an objective function based on physical arguments maximizing the probability of the pathway. We furthermore develop an automated pipeline to estimate the energy barriers of individual reactions in reaction networks. Combined, the methodology facilitates flexible and kinetically informed pathway investigations on large reaction networks by computational means, even for networks coming without kinetic annotation, such as those created via generative approaches for expanding molecular spaces. To demonstrate the methodology, we apply it on a chemical reaction network modeling the molecular space of glycolonitrile chemistry, where we search for pathways to glycine and glycolic acid using small molecules as precursors. I Introduction Kinetic studies of individual reactions have provided valuable insight into their mechanisms, which in turn can generate ways to optimize the reaction to suit the overall synthesis goals. However, molecules are often entangled in various concurrent reactions, which together form an interconnected web of reactions called a reaction network . Theoretical and computational modeling offers a possibility for analyzing these systems to various degrees of accuracy and computational efficiency, depending on the level of abstraction employed. In this work, we propose a method for integrating kinetics into the search for pathways in reaction networks modeled on a level of abstraction where the network is represented by a hypergraph and individual reactions are represented by directed hyperedges. The nodes of the hypergraph are molecular graphs (with nodes representing atoms and edges representing bonds) and hence the modeling is atomistically explicit. The work is a continuation of our previous work Pal et al. 2024 , which focused on the thermodynamics of individual reactions in the network. However, despite the name, thermodynamics does not help to elucidate the ‘dynamics’ of the system: how fast does the pathway in question run and how quickly does the system equilibrate? For example, under atmospheric conditions, the conversion of diamond to graphite is thermodynamically favorable. However, the energy barrier for this conversion is high, so, thankfully, for all practical purposes ‘diamonds are forever’. Therefore, kinetic parameters of reactions, such as the free energy barrier or the rate constant, might provide better alternatives for characterizing pathways. However, before one can use kinetic parameters to search for pathways in a reaction network, the network itself has to be constructed. This can be done in various ways. In one approach, the potential energy surface of a set of molecules is explored to elucidate trajectories with low-lying saddle points, separating the reactants from the products. Computationally intensive quantum mechanical calculations are used in most cases Dewyer et al. 2018 ; Zhao and Savoie 2021 ; Wang et al. 2014 , along with faster but less accurate semi-empirical methods in some Zhao et al. 2022 ; Woulfe and Savoie 2025 . Recently, generative diffusion models have also been used to perturb molecules to generate reaction networks with unknown mechanisms Duan et al. 2023 ; Kim et al. 2024 . A different approach is to use reaction templates for iteratively expanding the network from a set of starting molecules, possibly in conjunction with filters to exclude specific reactions Gao et al. 2016 ; Liu et al. 2021 ; Hoksza et al. 2014 . Further approaches include the generation of the reaction network of interest in a one-shot heuristical process by an expert Rappoport et al. 2014 ; Simm and Reiher 2017 ; Bergeler et al. 2015 , or the loading of an existing network from a database Wołos et al. 2020a ; Heald and Kroll 2020 . Except for the approach where the reaction network is generated by the exploration of the potential energy surface, individual reactions will need to be annotated with the kinetic parameters from a separate source, such as databases Jain et al. 2013 , estimations informed by experiments Wołos et al. 2020b ; Granda et al. 2018 , or dedicated quantum mechanical calculations McDermott et al. 2021 . Empirical relations, deduced via machine learning on available data on kinetics of reactions, have been used to estimate kinetic parameters for reactions Schaaf et al. 2023 ; Lewis-Atwell et al. 2022 ; Heinen et al. 2021 ; Stocker et al. 2022a ; Stocker et al. 2022b ; Wen et al. 2023 ; Chang et al. 2025 . However, these methods have limited success on ‘out of class’ molecules Vijay et al. 2022 . Once a reaction network has been constructed and the reactions annotated with appropriate kinetic data, one can employ diverse strategies to search and analyze pathways in it Kowalik et al. 2012 ; Robertson et al. 2020 ; Türtscher and Reiher 2023 . Previous work include Beam search based on suitable crafted scoring functions utilizing priority queues Grzybowski et al. 2023 , as well as Monte-Carlo search trees extended with a method to sample efficient pathways Lee et al. 2020 . Some analytical strategies have been used in the study of dynamics of pathways in reaction networks Shimada et al. 2024 ; van der Schaft et al. 2013 , but they are often limited by the availability of kinetic data and the scale of the networks. In this work, we develop an objective measure to rank pathways in a reaction network maximizing the probability of the entire pathway based on physical principles. Additionally, based on the notion that energy barriers of reactions are the determining factor for the kinetics of the network, we develop an automated methodology for computing such energy barriers, building on a combination of an array of existing computational tools. A key novelty is that in contrast to the standard manual intervention required to generate reaction trajectories from which the reaction barrier can be estimated, we try to automate this part of the process. Lastly, we formulate an integer linear program which allows the user to specify queries for pathways in the network in a very flexible and generic way. Solving this integer linear program returns all pathways fulfilling the criteria, ranked by the objective measure mentioned above. In short, this gives an automated, general, and flexible methodology for analyzing reaction networks, a methodology which offers a speed-up over manual search for pathways but with comparable quality. To exemplify its utility, we apply it to a reaction network expanded from the molecules glycolonitrile, ammonia and water, where we query for pathways to glycine, an amino acid, and glycolic acid, a molecule encountered in photorespiration in plants. II Definitions In this section, we present the mathematical frameworks used to model reaction networks and to specify searches for pathways in them, namely directed hypergraphs and integer hyperflows. This modeling is from Andersen et al. 2017 and Andersen et al. 2019 . II.1 Directed Hypergraphs A reaction network consists of a set of molecules, V