Pith. sign in

REVIEW 2 major objections 4 minor 50 references

Finding Thermodynamically Favorable Pathways in Chemical Reaction Networks Using Flows in Hypergraphs and Mixed-Integer Linear Programming

T0 review · 2 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A mixed-integer linear programming formulation of pathway search in chemical reaction networks enforces that every reaction in a returned pathway is thermodynamically favorable and ranks alternative pathways by cumulative Gibbs free…

desk verdict Useful thermodynamic extension of integer hyperflow pathway search, but Eq. (6) drops the RT/ln10 factor, so the reported numbers are computed with the wrong model. read the letter →

arxiv 2411.15900 v2 pith:AFOAW6PT submitted 2024-11-24 q-bio.MN cs.SYeess.SYphysics.chem-ph

Adittya Pal , Rolf Fagerberg , Jakob Lykke Andersen , Christoph Flamm , Peter Dittrich , Daniel Merkle This is my paper · ORCID
classification q-bio.MNcs.SYeess.SYphysics.chem-ph
keywords chemicalreactionnetworksintegerhyperflowsmixed-integerlinearprogrammingthermodynamicfeasibilityGibbsfreeenergyHCN-formamidechemistrypathwayenumerationoracle
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper enriches pathway search in chemical reaction networks with equilibrium thermodynamics. It formulates the search as a mixed-integer linear program over integer hyperflows, in which every reaction used by the returned pathway is forced to have a non-positive computed Gibbs free energy change at the chosen concentrations. The same objective function sums the free-energy changes along the pathway, so enumerated alternative pathways can be ranked by thermodynamic favorability. Applied to an HCN-formamide network grown from HCN, NH3, and H2O, the method reproduces the literature pathway but ranks it near the bottom of thirteen enumerated routes; the best-scoring route reaches the formamide trimer with objective value $-305.697$ kJ/mol, beating the literature route's $-171.926$ kJ/mol. If the method works as claimed, it gives experimental chemists a fast, thermodynamics-guided shortlist of synthesis pathways to test.

What carries the argument

The machinery combines integer hyperflows with continuous concentration variables in one mixed-integer linear program. A pathway is an integer flow vector $f_e$ on hyperedges representing reactions, extended with half-edges for sources and targets; flow conservation (2) fixes the overall reaction. Each molecule has a standard chemical potential $x^{G^0}_v$ from an external thermodynamic oracle and a log-concentration variable $x^K_v$ bounded by user-set limits (here $-3$ to $1$ in $\log_{10}$ M). The per-reaction free energy change is the linear expression $x^{\Delta G}_e = \sum_{v\in e^+} x^{G^0}_v - \sum_{v\in e^-} x^{G^0}_v + RT(\sum_{v\in e^+} x^K_v - \sum_{v\in e^-} x^K_v)$, so the concentration dependence is linear in log concentration. Binary variables $z_e$ indicate whether edge $e$ is used, linked to $f_e$ by big-M constraints; constraint (17) enforces $z_e = 1 \Rightarrow x^{\Delta G}_e \le 0$, and the objective (24) minimizes the sum of $x^{\Delta G}_e$ over used edges, interpreted by the authors as maximizing the probability of the pathway under an exponential ansatz. Optional integer variables $t_v$ enforce a partial temporal order on visited molecules, so a pathway can be read as a synthesis plan.

What would settle it

Fix the concentrations of HCN, NH3, and H2O at experimentally measured or reported reactor values inside the $10^{-3}$ to $10^1$ M window, remove the optimizer's freedom to move $x^K_v$, and re-solve the MILP for the formamide-trimer target. If the rank-0 pathway of Figure 4 becomes infeasible or contains a reaction with $x^{\Delta G}_e > 0$ at those fixed concentrations, the thermodynamic guarantee is an artifact of the concentration bounds rather than a property of the network.

Watch

Extended reading notes

Core claim

The central claim is that the implication constraint $f_e > 0 \Rightarrow x^{\Delta G}_e \le 0$, linearized as $x^{\Delta G}_e + M(z_e - 1) \le 0$, makes the MILP return only pathways in which every reaction with positive flow has a negative computed Gibbs free energy change, and that the objective $\min \sum_{e\in E} x^{\Delta G}_e$ gives a thermodynamically meaningful ranking of alternative pathways. The authors demonstrate this on a 67-vertex, 202-hyperedge HCN-formamide network queried from HCN, NH3, and H2O to the trimer of formamide. They enumerate thirteen distinct pathways; the top-ranked pathway has objective value $-305.697$ kJ/mol and uses five reactions and a tautomer not present in the earlier literature route, while the previously hypothesized pathway scores $-171.926$ kJ/mol. The flow-weighted net free energy difference is identical for every enumerated pathway, as expected for a path-independent quantity, so the ranking is driven by how the same overall driving force is distributed among the reactions.

Load-bearing premise

Everything rests on the assumption that the log-concentration variables are independent degrees of freedom within the hand-picked bounds from $-3$ to $1$ in $\log_{10}$ M; the solver always drives these variables to the bounds, so if realistic reactor concentrations fall outside that window, the reported 'thermodynamically favorable' pathways may not be favorable in the actual reactor.

Editorial extensions

If this is right

  • Pathway search can be screened so that only reactions with negative computed Gibbs free-energy changes enter the flow, eliminating thermodynamically implausible routes before any experimental work.
  • Because the solver enumerates over integer flow variables, users receive a ranked list of alternative pathways rather than a single solution, which hedges against errors in the chemical-potential estimates.
  • The thermodynamic oracle is modular: any method that maps a molecule to a chemical potential can replace the semi-empirical default, so the same filter and ranking can be rerun with more accurate potentials as they become available.
  • On the HCN-formamide network the enumeration yields thirteen pathways, and the literature route ranks only 12th under the chosen objective, with the top route using fewer reactions and a tautomer absent from the earlier study.
  • The optional temporal-ordering constraints make returned pathways interpretable as acyclic synthesis plans, at the cost of excluding pathways that contain catalytic cycles.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: because the optimizer always pushes every log-concentration variable to an extremal bound, the returned ranking should be read as what is achievable somewhere inside the chosen concentration window, not as favorability at any single realistic reactor state; rerunning with molecule-specific bounds would show how much the ranking depends on the window.
  • Editorial inference: the min-sum objective can favor pathways that load all driving force into one strongly exergonic step; the authors' suggested min-max variant, which ranks by the least-favorable reaction, is a natural next test and may select different routes.
  • Editorial inference: the same inequality template could become a kinetic filter; if a barrier-height oracle were available, replacing $x^{\Delta G}_e$ by an activation free energy in constraint (17) would return pathways whose rate-limiting steps remain accessible, coupling thermodynamics and kinetics in one search.
  • Editorial inference: enumeration over flow variables while leaving concentration variables free conflates pathway structure with the optimizer's freedom to choose concentrations; a two-stage procedure that fixes concentrations first and then enumerates would be a stricter test of whether a pathway is thermodynamically viable as written.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper extends the integer hyperflow framework for searching pathways in chemical reaction networks by adding thermodynamic constraints through a mixed-integer linear programming (MILP) formulation. Chemical potentials are assigned by a GFN2-xTB oracle, and log-concentration variables xK_v are introduced so that the Gibbs free energy change of each reaction can be expressed linearly. A constraint (5)/(17) is used to require that every reaction appearing in a returned pathway has a non-positive computed free energy change, and an objective function (24) ranks enumerated pathways by the sum of these free energy changes. The method is implemented in MØD and applied to an HCN-formamide reaction network, where 13 pathways to a formamide trimer are enumerated and compared with a previously reported route.

Significance. If the technical issues are corrected, the work provides a useful and modular extension of an existing pathway-search tool: it demonstrates how an external thermodynamic oracle can be combined with MILP-based enumeration over integral hyperflows, and the case study gives a concrete proof of concept. The authors make the implementation and generated data publicly available, which is a strength. The central mathematical idea is not circular: it applies standard equilibrium thermodynamics and an external quantum-chemical oracle. However, the current numerical results and the associated claim that the returned pathways are thermodynamically favorable rest on an incorrect linearization of the concentration term, and the free concentration variables make the thermodynamic filter sensitive to arbitrarily chosen bounds; these issues must be resolved before the quantitative claims can be accepted.

major comments (2)
  1. [§3.1.1, Eq. (6)] Equation (6) is inconsistent with the derivation that immediately follows it. The paper defines xK_v as log10(conc(v)), so the concentration term in the linearized expression for x∆G_e must carry the factor RT/ln(10), not RT. As written, RT·(Σ xK out − Σ xK in) overestimates the concentration correction by a factor of about 2.303. This error propagates into constraint (17), the objective function (24), and all reported reaction free energies in Figures 3–4 and Table 6. For example, for hyperedge e1 in Figure 3, using the displayed xG0 values and ΔxK = −1, the incorrect formula gives about −94.4 kJ/mol, matching the reported −94.141 kJ/mol, whereas the correct formula gives about −93.0 kJ/mol. The authors need to correct Eq. (6), rerun the MILP enumeration, and re-evaluate the pathway rankings; the current numbers and the claim that the listed pathways are thermodynamically favorable under the stated concentration model are not reliable.
  2. [§4, Tables 3c/4c and accompanying text] The authors observe that the optimizer always assigns the log-concentration variables xK_v to the extremal values of the hand-picked uniform bounds (−3 and 1). This means that the thermodynamic filter in (5)/(17) does not evaluate whether a pathway is favorable at a specified, physically meaningful reactor composition; instead, it optimizes over the concentration variables to make the pathway favorable. The paper acknowledges this and offers a reinterpretation in terms of perturbation directions, but the abstract and introduction still state that the search is constrained to return pathways containing only thermodynamically favorable reactions. This overstates what the method computes. The authors should state this limitation more prominently and, ideally, report sensitivity of the enumerated rankings to the chosen bounds, for example by repeating the enumeration with different xK ranges or with species-specific bounds.
minor comments (4)
  1. [§3.1.2, Eq. (8)] Equation (8) is written with a “max” operator, but the surrounding text describes minimization and Eq. (24) uses “min”. Please reconcile these; the intended objective should be stated unambiguously.
  2. [§4, Table 6] The statement that the last column is path-independent and therefore the same for all pathways is not exactly reflected in the table, as the reported values range from −399.835 to −399.837 kJ/mol. A brief explanation of numerical tolerance would help.
  3. [§3.1.1] There is a typo: “the absolute temperate of the system” should be “the absolute temperature of the system.”
  4. [§4, Tables 3c and 4c] The column headers “x∆G_v” in these tables are confusing because the listed values are standard chemical potentials xG0_v in atomic units, not free energy differences. Rename the headers to “xG0_v”.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the MILP search is a constraint-based optimization over an external thermodynamic oracle, with no fitted input renamed as a prediction.

full rationale

The derivation chain is self-contained. The paper extends the integer hyperflow formalism of Andersen et al. (2019) with standard equilibrium thermodynamics and chemical potentials from GFN2-xTB, an external quantum-chemical oracle. The thermodynamic filter in constraint (5), linearized in (17), literally enforces xΔG_e ≤ 0 for every used reaction, so the statement that returned pathways contain only thermodynamically favorable reactions is a direct consequence of the constraint, not a hidden input or a fitted target. The objective function (8)/(24) ranks pathways by the same xΔG values; ranking by one's own explicitly chosen objective is not circular, and the paper labels it as 'our chosen objective function.' The enumeration includes the literature pathway from Kua & Thrush (2016) as a sanity check and reports better-scoring alternatives, which is an external benchmark rather than a fitted target. The self-citations to Andersen et al. (2019), Fagerberg et al. (2018), and Banke et al. (2023) support the underlying hyperflow, synthesis-plan, and realizability machinery; these are code-implemented (MØD) and independent of the thermodynamic claims, so they do not raise the circularity score. No uniqueness theorem or ansatz is imported from the authors' prior work to force the present conclusions. The separately noted concern about a possible missing 1/ln(10) factor in Eq. (6) is a substantive modeling and correctness issue, but it is not circularity: it does not make the output equivalent to the input by construction. No circular step is present.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central model rests on standard equilibrium thermodynamics and an external semi-empirical oracle, with hand-picked concentration bounds and a big-M constant as the main free parameters. No new physical entities are postulated. The near-equilibrium and Evans-Polanyi-Semenov assumptions are explicit but unquantified.

free parameters (2)
  • concentration bounds xK_min=-3, xK_max=1 = -3 to 1 (log10 M)
    Hand-chosen uniform bounds on log10 concentration for all molecules (Appendix B); the optimizer pins xK_v to these bounds, so the thermodynamic feasibility region and resulting pathway ranking depend directly on this choice.
  • Big-M constant M = 100
    Chosen as a sufficiently large constant for the big-M linearization; the authors state M=100 suffices but do not derive it from the model's variable ranges.
assumptions (4)
  • domain assumption The reaction network is near equilibrium, so equilibrium thermodynamics can be applied as a first-order approximation for pathway driving forces.
    Section 3 states the system is perturbed only slightly from equilibrium; this justifies using ΔG constraints on individual reactions despite the system being driven.
  • domain assumption Reaction favorability is a valid proxy for reaction kinetics (Evans-Polanyi-Semenov principle).
    Section 3.4 invokes this principle to justify neglecting activation barriers and using only thermodynamic driving forces.
  • domain assumption GFN2-xTB-calculated chemical potentials under standard conditions are accurate enough for the pathway search.
    Section 3.4 and Appendix A use GFN2-xTB as the thermodynamic oracle; no uncertainty quantification is provided.
  • standard math The integer hyperflow model from (Andersen et al., 2019) correctly represents pathway stoichiometry.
    Section 2 inherits the hyperflow formalism; the paper does not re-derive it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Finding Thermodynamically Favorable Pathways in Chemical Reaction Networks Using Flows in Hypergraphs and Mixed-Integer Linear Programming." pith.science (2026). https://pith.science/paper/AFOAW6PT

@misc{pith2026241115900,
  author       = {Pith},
  title        = {Pith review of: Finding Thermodynamically Favorable Pathways in Chemical Reaction Networks Using Flows in Hypergraphs and Mixed-Integer Linear Programming},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AFOAW6PT}},
  note         = {Machine review of arXiv:2411.15900}
}
read the original abstract

The search for pathways that optimize the formation of a particular target molecule in a reaction network is a key problem in many settings, including reactor systems. Chemical reaction networks are mathematically well represented as hypergraphs, modeling that facilitates the search for pathways by computational means. We propose to enrich an existing search method for pathways by including thermodynamic principles. In more detail, we give a mixed-integer linear programming (mixed ILP) formulation of the search problem into which we integrate chemical potentials and concentrations for individual molecules, enabling us to constrain the search to return pathways containing only thermodynamically favorable reactions. Moreover, if multiple possible pathways are found, we can rank these by objective functions based on thermodynamics. As an example of use, we apply the framework to a reaction network representing the HCN-formamide chemistry. Alternative pathways to the one currently hypothesized in the literature are queried and enumerated, including some that score better according to our chosen objective function.

Figures

Figures reproduced from arXiv: 2411.15900 by the authors.

Figure 1
Figure 1. A directed hypergraph 1a, where circles are vertices and the squares are hyperedges. The other three subfigures 1b, 1c, and 1d show possible integer hyperflows in the hypergraph, each representing a pathway, where hyperedges with non-zero flow are drawn in bold. The inflow and the outflow are depicted as arrows in and out of the source and target vertices of the pathway. Hyperedges with multiple copies of a vertex a… view at source ↗
Figure 2
Figure 2. A representation of the hypergraph from [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. 3a: The pathway with the sum of free energies −171.926 kJ mol−1 and Table 3b: the molecules used with the corresponding number in the earlier work (Kua & Thrush, 2016). Table 3c and Table 3d: The MILP solution with the x ∆G v values are in atomic units (from xTB) while the x ∆G e are in kJ mol−1 . The x K v values are the dimensionless equivalent of log10 [concentration] 1 mol lit−1 . multiple times, making seven re… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: 4a: The pathway with minimum sum of free energies (−305.697 kJ mol−1 ) returned as the optimal by our model and Table 4b: the molecules used with the corresponding number in the earlier work. Table 4c and Table 4d: MILP solution and the assignment of the MILP variables…
Figure 5
Figure 5. Figure 5: Graph transformation rules (or reaction templates) used to generate the reaction network 6 in which pathways were queried. The reverse of rules 5b, 5c and 5d was used, while the reverse of rule 5a was not used while expanding the hypergraph. N CH g0 NH3 g1 H2O g2 H2N N…
Figure 6
Figure 6. Figure 6: Hypergraph with |V | = 67 and |E| = 202, representing the reaction network considered in this work, provided here for the sake of completeness. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 31 canonical work pages

  1. [1]

    https://github.com/AdittyaPal/thermodynamicsWithILP.git, 2024

    supplementary data repository on G it H ub. https://github.com/AdittyaPal/thermodynamicsWithILP.git, 2024

    work page 2024

  2. [2]

    http://mod.imada.sdu.dk, 2024

    M D . http://mod.imada.sdu.dk, 2024

    work page 2024

  3. [3]

    https://github.com/openbabel/openbabel, 2024

    O pen B abel G ithub repository. https://github.com/openbabel/openbabel, 2024

    work page 2024

  4. [4]

    https://github.com/AdittyaPal/thermodynamicsWithILP/blob/5f7aed70fe8706bbe0ffc2efedd68316d44e0f42/listOfSolutions.pdf, 2025

    list of solutions returned by the solver. https://github.com/AdittyaPal/thermodynamicsWithILP/blob/5f7aed70fe8706bbe0ffc2efedd68316d44e0f42/listOfSolutions.pdf, 2025

    work page 2025

  5. [5]

    Andersen, J., Flamm, C., Merkle, D., and Stadler, P. F. Maximizing output and recognizing autocatalysis in chemical reaction networks is N P -complete . J Syst Chem, 3 0 (1), 01 2012. doi:10.1186/1759-2208-3-1. URL https://doi.org/10.1186/1759-2208-3-1

    work page doi:10.1186/1759-2208-3-1 2012

  6. [6]

    L., Flamm, C., Merkle, D., and Stadler, P

    Andersen, J. L., Flamm, C., Merkle, D., and Stadler, P. F. A S oftware P ackage for C hemically I nspired G raph T ransformation. In Graph Transformation, pp.\ 73--88, Cham, 2016. Springer International Publishing. URL https://doi.org/10.1007/978-3-319-40530-8_5

    doi:10.1007/978-3-319-40530-8_5 2016

  7. [7]

    L., Flamm, C., Merkle, D., and Stadler, P

    Andersen, J. L., Flamm, C., Merkle, D., and Stadler, P. F. Chemical T ransformation M otifs— M odelling P athways as I nteger H yperflows. IEEE/ACM Trans. Comput. Biol. Bioinformatics, 16 0 (2): 0 510–523, April 2019. ISSN 1545-5963. doi:10.1109/TCBB.2017.2781724. URL https://doi.org/10.1109/TCBB.2017.2781724

    arXiv 2019

  8. [8]

    Circuit theory for C hemical R eaction N etworks

    Avanzini, F., Freitas, N., and Esposito, M. Circuit theory for C hemical R eaction N etworks. Phys. Rev. X, 13: 0 021041, Jun 2023. doi:10.1103/PhysRevX.13.021041. URL https://doi.org/10.1103/PhysRevX.13.021041

    doi:10.1103/physrevx.13.021041 2023

Show all 50 references
  1. [9]

    L., Flamm, C., Merkle, D., and Stadler, P

    Banke, S., Fagerberg, R., Andersen, J. L., Flamm, C., Merkle, D., and Stadler, P. F. On the R ealisability of C hemical P athways, 2023. URL https://doi.org/10.1007/978-981-99-7074-2_32

    2023 doi

  2. [10]

    Extended tight-binding quantum chemistry methods

    Bannwarth, C., Caldeweyher, E., Ehlert, S., Hansen, A., Pracht, P., Seibert, J., Spicher, S., and Grimme, S. Extended tight-binding quantum chemistry methods. WIREs Computational Molecular Science, 11 0 (2): 0 e1493, 2021. doi:10.1002/wcms.1493. URL https://doi.org/10.1002/wcms.1493

    2021 doi

  3. [11]

    W., Redkar, N

    Barter, D., Clark Spotte-Smith, E. W., Redkar, N. S., Khanwale, A., Dwaraknath, S., Persson, K. A., and Blau, S. M. Predictive stochastic analysis of massive filter-based electrochemical reaction networks. Digital Discovery, 2: 0 123--137, 2023. doi:10.1039/D2DD00117A. URL htt...

    2023 doi

  4. [12]

    Beasley, J. E. and Planes, F. J. Recovering metabolic pathways via optimization. Bioinformatics, 23 0 (1): 0 92--98, 10 2006. ISSN 1367-4803. doi:10.1093/bioinformatics/btl554. URL https://doi.org/10.1093/bioinformatics/btl554

    2006 doi

  5. [13]

    M., Patel, H

    Blau, S. M., Patel, H. D., Spotte-Smith, E. W. C., Xie, X., Dwaraknath, S., and Persson, K. A. A chemically consistent graph architecture for massive reaction networks applied to solid-electrolyte interphase formation. Chem. Sci., 12: 0 4931--4939, 2021. doi:10.1039/D0SC05647B...

    2021 doi

  6. [14]

    J., Segler, M

    Bradshaw, J., Paige, B., Kusner, M. J., Segler, M. H. S., and Hernández-Lobato, J. M. Barking up the right tree: an approach to search over molecule synthesis D A G s, 2020. URL https://arxiv.org/abs/2012.11522

    2020 arXiv

  7. [15]

    Cantú, A. G. and Nicolis, G. T oward a T hermodynamic C haracterization of C hemical R eaction N etworks. Journal of Non-Equilibrium Thermodynamics, 31 0 (1): 0 23--46, 2006. doi:10.1515/JNETDY.2006.003. URL https://doi.org/10.1515/JNETDY.2006.003

    2006 doi

  8. [16]

    and Fiaschi, L

    Cococcioni, M. and Fiaschi, L. The B ig- M method with the numerical infinite M . M. Optim Lett, 15: 0 2455--2468, September 2020. doi:10.1007/s11590-020-01644-6. URL https://doi.org/10.1007/s11590-020-01644-6

    2020 doi

  9. [17]

    G., Johnson, M

    Dana, A. G., Johnson, M. S., Allen, J. W., et al. Automated reaction kinetics and network exploration ( A rkane): A statistical mechanics, thermodynamics, transition state theory, and master equation software. International Journal of Chemical Kinetics, 55 0 (6): 0 300--323, 2...

    2023 doi

  10. [18]

    Fagerberg, R., Flamm, C., Kianian, R., Merkle, D., and Stadler, P. F. Finding the K best synthesis plans. J Cheminform, 10 0 (19), April 2018. doi:10.1186/s13321-018-0273-z. URL https://doi.org/10.1186/s13321-018-0273-z

    2018 doi

  11. [19]

    coin-or/cbc: Release releases/2.10.12, August 2024

    Forrest, J., Ralphs, T., Vigerske, S., et al. coin-or/cbc: Release releases/2.10.12, August 2024. URL https://doi.org/10.5281/zenodo.13347261

    2024 doi

  12. [20]

    S., Broadbelt, L

    Henry, C. S., Broadbelt, L. J., and Hatzimanikatis, V. Thermodynamics- B ased M etabolic F lux A nalysis. Biophysical Journal, 92 0 (5): 0 1792--1805, 2007. ISSN 0006-3495. doi:10.1529/biophysj.106.093138. URL https://doi.org/10.1529/biophysj.106.093138

    2007 doi

  13. [21]

    Analysis of mass-action systems by split network translation

    Johnston, M. Analysis of mass-action systems by split network translation. J Math Chem, 60: 0 195–218, November 2021. doi:10.1007/s10910-021-01299-3. URL https://doi.org/10.1007/s10910-021-01299-3

    2021 doi

  14. [22]

    A linear programming approach to weak reversibility and linear conjugacy of chemical reaction networks

    Johnston, M., Siegel, D., and Szederkényi, G. A linear programming approach to weak reversibility and linear conjugacy of chemical reaction networks. J Math Chem, 50: 0 274–288, September 2011. doi:10.1007/s10910-011-9911-7. URL https://doi.org/10.1007/s10910-011-9911-7

    2011 doi

  15. [23]

    Karp, R. M. Reducibility among combinatorial problems. In Complexity of Computer Computations, pp.\ 85--103, NY, 1972. Plenum Press. URL https://doi.org/10.1007/978-1-4684-2001-2_9

    1972 doi

  16. [24]

    and Thrush, K

    Kua, J. and Thrush, K. L. H C N , F ormamidic A cid, and F ormamide in A queous S olution: A F ree- E nergy M ap. The Journal of Physical Chemistry B, 120 0 (33): 0 8175--8185, 2016. doi:10.1021/acs.jpcb.6b01690. URL https://doi.org/10.1021/acs.jpcb.6b01690

    2016 doi

  17. [25]

    and Jiang, X

    Li, J. and Jiang, X. M ol- B E R T : A n E ffective M olecular R epresentation with B E R T for M olecular P roperty prediction. Wireless Communications and Mobile Computing, 2021 0 (1): 0 7181815, 2021. doi:10.1155/2021/7181815. URL https://doi.org/10.1155/2021/7181815

    2021 doi

  18. [26]

    S., et al

    Liu, M., Grinberg Dana, A., Johnson, M. S., et al. R eaction M echanism G enerator v3.0: A dvances in A utomatic M echanism G eneration. Journal of Chemical Information and Modeling, 61 0 (6): 0 2686--2696, 2021. doi:10.1021/acs.jcim.0c01480. URL https://doi.org/10.1021/acs.jc...

    2021 doi

  19. [27]

    The common optimization interface for operations research: Promoting open-source software in the operations research community

    Lougee-Heimer, R. The common optimization interface for operations research: Promoting open-source software in the operations research community. IBM J. Res. Dev., 47 0 (1): 0 57–66, January 2003. ISSN 0018-8646. doi:10.1147/rd.471.0057. URL https://doi.org/10.1147/rd.471.0057

    2003 doi

  20. [28]

    A uto M e K in2021: An open-source program for automated reaction discovery

    Martínez-Núñez, Emilio, Barnes, L., G., et al. A uto M e K in2021: An open-source program for automated reaction discovery. Journal of Computational Chemistry, 42 0 (28): 0 2036--2048, 2021. doi:10.1002/jcc.26734. URL https://doi.org/10.1002/jcc.26734

    2021 doi

  21. [29]

    J., Dwaraknath, S

    McDermott, M. J., Dwaraknath, S. S., and Persson, K. A. A graph-based network for predicting chemical reaction pathways in solid-state materials synthesis. Nat Commun, 12, April 2021. doi:10.1038/s41467-021-23339-x. URL https://doi.org/10.1038/s41467-021-23339-x

    2021 doi

  22. [30]

    and Castellana, M

    Miangolarra, A. and Castellana, M. On N on-ideal C hemical R eaction N etworks and P hase S eparation. J Stat Phys, 190 0 (23), November 2023. doi:10.1007/s10955-022-03037-8. URL https://doi.org/10.1007/s10955-022-03037-8

    2023 doi

  23. [31]

    Machine learning for molecular simulation

    Noé, F., Tkatchenko, A., Müller, K.-R., and Clementi, C. Machine learning for molecular simulation. Annual Review of Physical Chemistry, 71 0 (Volume 71, 2020): 0 361--390, 2020. ISSN 1545-1593. doi:10.1146/annurev-physchem-042018-052331. URL https://doi.org/10.1146/annurev-ph...

    2020 doi

  24. [32]

    A., Mucelini, J., Soares, M

    Pinheiro, G. A., Mucelini, J., Soares, M. D., Prati, R. C., Da Silva, J. L. F., and Quiles, M. G. Machine learning prediction of nine molecular properties based on the S M I L E S representation of the Q M 9 quantum-chemistry dataset. The Journal of Physical Chemistry A, 124 0...

    2020 doi

  25. [33]

    Planes, F. J. and Beasley, J. E. A critical examination of stoichiometric and path-finding approaches to metabolic pathways. Briefings in Bioinformatics, 9 0 (5): 0 422--436, 04 2008. ISSN 1467-5463. doi:10.1093/bib/bbn018. URL https://doi.org/10.1093/bib/bbn018

    2008 doi

  26. [34]

    and Gagliardi, A

    Rahaman, O. and Gagliardi, A. Deep learning total energies and orbital energies of large organic molecules using hybridization of molecular fingerprints. Journal of Chemical Information and Modeling, 60 0 (12): 0 5971--5983, 2020. doi:10.1021/acs.jcim.0c00687. URL https://doi....

    2020 doi

  27. [35]

    and Esposito, M

    Rao, R. and Esposito, M. N onequilibrium T hermodynamics of C hemical R eaction N etworks: W isdom from S tochastic T hermodynamics. Phys. Rev. X, 6: 0 041064, Dec 2016. doi:10.1103/PhysRevX.6.041064. URL https://doi.org/10.1103/PhysRevX.6.041064

    2016 doi

  28. [36]

    K., Casewit, C

    Rappe, A. K., Casewit, C. J., Colwell, K. S., Goddard, W. A. I., and Skiff, W. M. U F F , a full periodic table force field for molecular mechanics and molecular dynamics simulations. Journal of the American Chemical Society, 114 0 (25): 0 10024--10035, 1992. doi:10.1021/ja000...

    1992 doi

  29. [37]

    C., and Hatzimanikatis, V

    Salvy, P., Fengos, G., Ataman, M., Pathier, T., Soh, K. C., and Hatzimanikatis, V. py T F A and mat T F A : a P ython package and a M atlab toolbox for T hermodynamics-based F lux A nalysis. Bioinformatics, 35 0 (1): 0 167--169, 07 2018. ISSN 1367-4803. doi:10.1093/bioinformat...

    2018 doi

  30. [38]

    and Seifert, U

    Schmiedl, T. and Seifert, U. Stochastic thermodynamics of chemical reaction networks . The Journal of Chemical Physics, 126 0 (4): 0 044101, 01 2007. ISSN 0021-9606. doi:10.1063/1.2428297. URL https://doi.org/10.1063/1.2428297

    2007 doi

  31. [39]

    M., Strutz, J., Broadbelt, L

    Shebek, K. M., Strutz, J., Broadbelt, L. J., and Tyo, K. E. J. Pickaxe: a P ython library for the prediction of novel metabolic reactions. BMC Bioinformatics, 24 0 (106), March 2023. doi:10.1186/s12859-023-05149-8. URL https://doi.org/10.1186/s12859-023-05149-8

    2023 doi

  32. [40]

    and Hangos, K

    Szederkényi, G. and Hangos, K. Finding complex balanced and detailed balanced realizations of chemical reaction networks. J Math Chem, 49: 0 1163–1179, March 2011. doi:10.1007/s10910-011-9804-9. URL https://doi.org/10.1007/s10910-011-9804-9

    2011 doi

  33. [41]

    Quantum mechanics and machine learning synergies: Graph attention neural networks to predict chemical reactivity

    Tavakoli, M., Mood, A., Van Vranken, D., and Baldi, P. Quantum mechanics and machine learning synergies: Graph attention neural networks to predict chemical reactivity. Journal of Chemical Information and Modeling, 62 0 (9): 0 2121--2132, 2022. doi:10.1021/acs.jcim.1c01400. UR...

    2022 doi

  34. [42]

    Exposing the limitations of molecular machine learning with activity cliffs

    van Tilborg, D., Alenicheva, A., and Grisoni, F. Exposing the limitations of molecular machine learning with activity cliffs. Journal of Chemical Information and Modeling, 62 0 (23): 0 5938--5951, 2022. doi:10.1021/acs.jcim.2c01073. URL https://doi.org/10.1021/acs.jcim.2c01073

    2022 doi

  35. [43]

    Challenges for machine learning force fields in reproducing potential energy surfaces of flexible molecules

    Vassilev-Galindo, V., Fonseca, G., Poltavsky, I., and Tkatchenko, A. Challenges for machine learning force fields in reproducing potential energy surfaces of flexible molecules. The Journal of Chemical Physics, 154 0 (9): 0 094119, 03 2021. ISSN 0021-9606. doi:10.1063/5.003851...

    2021 doi

  36. [44]

    Vijay, S., Kastlunger, G., Chan, K., and Nørskov, J. K. Limits to scaling relations between adsorption energies? The Journal of Chemical Physics, 156 0 (23): 0 231102, 06 2022. ISSN 0021-9606. doi:10.1063/5.0096625. URL https://doi.org/10.1063/5.0096625

    2022 doi

  37. [45]

    Willis, M. J. and von Stosch, M. Inference of chemical reaction networks using mixed integer linear programming. Computers & Chemical Engineering, 90: 0 31--43, 2016. ISSN 0098-1354. doi:10.1016/j.compchemeng.2016.04.019. URL https://doi.org/10.1016/j.compchemeng.2016.04.019

    2016 doi

  38. [46]

    and Hutchison, G

    Yoshikawa, N. and Hutchison, G. Fast, efficient fragment-based coordinate generation for O pen B abel. J Cheminform, 11 0 (49), August 2019. doi:10.1186/s13321-019-0372-5. URL https://doi.org/10.1186/s13321-019-0372-5

    2019 doi

  39. [47]

    A., Silcock, J

    Young, T. A., Silcock, J. J., Sterling, A. J., and Duarte, F. autod E : A utomated C alculation of R eaction E nergy P rofiles— A pplication to O rganic and O rganometallic R eactions. Angewandte Chemie International Edition, 60 0 (8): 0 4266--4274, 2021. doi:10.1002/anie.2020...

    2021 doi

  40. [48]

    Zichi, L., Barter, D., Sivonxay, E., Spotte-Smith, E. W. C., Mohanakrishnan, R. S., Chan, E. M., Persson, K. A., and Blau, S. M. R N M C : kinetic M onte C arlo implementations for complex reaction networks. Journal of Open Source Software, 9 0 (104): 0 7244, 2024. doi:10.2110...

    2024 doi

  41. [49]

    S., Nebgen, B

    Zubatyuk, R., Smith, J. S., Nebgen, B. T., Tretiak, S., and Isayev, O. Teaching a neural network to attach and detach electrons from molecules. Nat Commun, 12, August 2021. doi:10.1038/s41467-021-24904-0. URL https://doi.org/10.1038/s41467-021-24904-0

    2021 doi

  42. [50]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.