CoChat

Catalyst: Fast and flexible modeling of reaction networks

Open access
Loman, Torkel E., Ma, Yingbo, Ilin, Vasily, Gowda, Shashi, Korsbo, Niklas, Yewale, Nikhil, Rackauckas, Chris, Isaacson, Samuel A. · 2023 · PLoS Computational Biology

Published version of a preprint

The DOI matches its Crossref record; 80 of the 80 references it cites were found in a registry or an index; 11 of the 26 sources its claims cite were checked for support.

Abstract

We introduce Catalyst.jl, a flexible and feature-filled Julia library for modeling and high-performance simulation of chemical reaction networks (CRNs). Catalyst supports simulating stochastic chemical kinetics (jump process), chemical Langevin equation (stochastic differential equation), and reaction rate equation (ordinary differential equation) representations for CRNs. Through comprehensive benchmarks, we demonstrate that Catalyst simulation runtimes are often one to two orders of magnitude faster than other popular tools. More broadly, Catalyst acts as both a domain-specific language and an intermediate representation for symbolically encoding CRN models as Julia-native objects. This enables a pipeline of symbolically specifying, analyzing, and modifying CRNs; converting Catalyst models to symbolic representations of concrete mathematical models; and generating compiled code for numerical solvers. Leveraging ModelingToolkit.jl and Symbolics.jl, Catalyst models can be analyzed, simplified, and compiled into optimized representations for use in numerical solvers. Finally, we demonstrate Catalyst's broad extensibility and composability by highlighting how it can compose with a variety of Julia libraries, and how existing open-source biological modeling projects have extended its intermediate representation.

Overview

CoChat Summary

## TL;DR

Catalyst.jl is a Julia library for modeling and simulating chemical reaction networks that combines a symbolic intermediate representation, a DSL for reaction notation, and high-performance numerical solvers. In benchmarks across five CRN models, it was often one to two orders of magnitude faster than other popular tools.

## Key findings

  • Catalyst supports three CRN representations: stochastic chemical kinetics via jump processes, the chemical Langevin equation (SDE), and reaction rate equations (ODEs).
  • In benchmarks on five CRN models, Catalyst “typically outperforms the other packages, often by at least an order of magnitude,” and the abstract says runtimes are often “one to two orders of magnitude faster” than other popular tools.
  • Catalyst’s symbolic representation enables automatic Jacobian computation, elimination of conservation laws, simplification of symbolic DAE models, and network analysis including linkage classes, weak reversibility, and deficiency indices.
  • Catalyst is designed to compose with other Julia packages; the paper gives examples including parameter fitting, inference, identifiability analysis, sensitivity analysis, steady-state analysis, bifurcation analysis, moment closure, finite-state projection, and delay SSA tooling.

## Methods

  • The paper introduces Catalyst.jl as a Julia-based DSL and symbolic IR for CRNs, using ModelingToolkit.jl and Symbolics.jl to encode, analyze, simplify, and compile models.
  • Benchmarking compared Catalyst against BioNetGen, COPASI, GillesPy2, and Matlab SimBiology using five CRN models from prior work; ODE and SSA simulations were run, and SDEs were not benchmarked because fewer packages supported them.
  • Performance analysis used multiple solver/method combinations and options (e.g., lsoda, CVODE, Julia-native solvers, Jacobian representations, linear solvers, and SSA methods), with median runtimes over several runs.

## Limitations

  • The paper states that Catalyst does not currently support hybrid methods or τ-leaping-based solvers, though these are planned for future updates.
  • The paper also notes planned future support for spatial models, reaction-diffusion master equation solvers, and continuous-space transport models.
  • Evident from the text: the reported comparisons are limited to five CRN models and to the specific tools/methods that were available and benchmarked; for some largest models, the slowest tools and methods were not benchmarked because runs longer than 4 days were not permitted.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1371/journal.pcbi.1011530

Cites

80 of 80 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig 1 Catalyst connects an intuitive domain-specific language with a well-supported intermediate representation. The extracellular signal-regulated kinase (ERK) network is important to the regulation of many cellular functions, and its disruption has been implicated in cancer [35]. (a) A CRN representation of the ERK network. (b) A model of the ERK CRN can be implemented in Julia through the Catalyst DSL, using code very similar to the actual CRN representation. (c) From this code, the DSL generates a ReactionSystem intermediate representation (IR) that is the target structure for a range of supported simulation and analysis methods.
Fig 1 Catalyst connects an intuitive domain-specific language with a well-supported intermediate representation. The extracellular signal-regulated kinase (ERK) network is important to the regulation of many cellular functions, and its disruption has been implicated in cancer [35]. (a) A CRN representation of the ERK network. (b) A model of the ERK CRN can be implemented in Julia through the Catalyst DSL, using code very similar to the actual CRN representation. (c) From this code, the DSL generates a ReactionSystem intermediate representation (IR) that is the target structure for a range of supported simulation and analysis methods.CC BY · PubMed Central
Fig 2 Catalyst models can be simulated using both deterministic and stochastic interpretations. (a) The Brusselator network contains two species (X and Y) and two parameters (A and B, in practical implementation these are species present in excess, but they can in practice be considered parameters) [39, 40]. Here, we show the four reactions of the Brusselator CRN, and its implementation using the Catalyst DSL. (b-d) Simulations of models for the Brusselator at the three physical scales supported by Catalyst (RRE, CLE, SSA). Post-processing has been carried out on the plots to improve their visualization in this article’s format. (b) While B > 1 + A2, the deterministic model exhibits a limit cycle. This is confirmed using ODE RRE simulations. (c) The model can also be simulated using the stochastic CLE interpretation. (d) Finally, the discrete, stochastic, jump process interpretation is simulated via Gillespie’s direct method. The system displays a limit cycle even though B < 1 + A2, confirming the well known phenomenon of noise induced oscillations [41].
Fig 2 Catalyst models can be simulated using both deterministic and stochastic interpretations. (a) The Brusselator network contains two species (X and Y) and two parameters (A and B, in practical implementation these are species present in excess, but they can in practice be considered parameters) [39, 40]. Here, we show the four reactions of the Brusselator CRN, and its implementation using the Catalyst DSL. (b-d) Simulations of models for the Brusselator at the three physical scales supported by Catalyst (RRE, CLE, SSA). Post-processing has been carried out on the plots to improve their visualization in this article’s format. (b) While B > 1 + A2, the deterministic model exhibits a limit cycle. This is confirmed using ODE RRE simulations. (c) The model can also be simulated using the stochastic CLE interpretation. (d) Finally, the discrete, stochastic, jump process interpretation is simulated via Gillespie’s direct method. The system displays a limit cycle even though B < 1 + A2, confirming the well known phenomenon of noise induced oscillations [41].CC BY · PubMed Central
Fig 3 Simulations of Catalyst models outperform those of other modeling packages. Benchmarks of simulation runtimes for Catalyst and four other modeling packages (BioNetGen, COPASI, GillesPy2, and Matlab SimBiology). The benchmarks were run on the multi-state (Multistate, 9 species and 18 reactions [47]), multi-site (Multisite 2, 66 species and 288 reactions [48]), epidermal growth factor receptor signalling (Egfr_net, 356 species and 3749 reactions [49]), B-cell receptor (1122 species and 24388 reactions [50]), and high-affinity human IgE receptor signalling (Fceri_gamma2, 3744 species and 58276 reactions [51]) models. (a-e) Benchmarks of deterministic RRE ODE simulations of the five models. Each bar shows, for a given method, the runtime to simulate the model (to steady-state for those that approach a steady-state). For Catalyst, we show the three best-performing native Julia methods, as well as the performance of lsoda and CVODE. For each of the other tools, we show its best-performing method. Identical values for absolute and relative tolerance are used across all packages and methods. For each benchmark, the method options used can be found in Section 4.1, the exact benchmark times in Table A in S1 Text, and further details on the solver options for each tool in Section B in S1 Text. While this figure only contains the most performant methods, a full list of methods investigated can be found in Section B in S1 Text, with their results described in Figs A and B in S1 Text. (f-j) Benchmarks of stochastic chemical kinetics SSA simulations of the five models. Via JumpProcesses.jl, Catalyst can use several different algorithms (e.g. Direct, Sorting Direct, RSSA, and RSSACR above) for exact Gillespie simulations. Here, the simulation runtime is plotted against the (physical) final time of the simulation. Due to their long runtimes, some tools were not benchmarked for the largest models. We note that, in [52], it was remarked that BioNetGen (dashed orange lines) use a pseudo-random number generator in SSAs that, while fast, is of lower quality than many (slower) modern generators such as Mersenne Twister. For full details on benchmarks, see Section 4.1.
Fig 3 Simulations of Catalyst models outperform those of other modeling packages. Benchmarks of simulation runtimes for Catalyst and four other modeling packages (BioNetGen, COPASI, GillesPy2, and Matlab SimBiology). The benchmarks were run on the multi-state (Multistate, 9 species and 18 reactions [47]), multi-site (Multisite 2, 66 species and 288 reactions [48]), epidermal growth factor receptor signalling (Egfr_net, 356 species and 3749 reactions [49]), B-cell receptor (1122 species and 24388 reactions [50]), and high-affinity human IgE receptor signalling (Fceri_gamma2, 3744 species and 58276 reactions [51]) models. (a-e) Benchmarks of deterministic RRE ODE simulations of the five models. Each bar shows, for a given method, the runtime to simulate the model (to steady-state for those that approach a steady-state). For Catalyst, we show the three best-performing native Julia methods, as well as the performance of lsoda and CVODE. For each of the other tools, we show its best-performing method. Identical values for absolute and relative tolerance are used across all packages and methods. For each benchmark, the method options used can be found in Section 4.1, the exact benchmark times in Table A in S1 Text, and further details on the solver options for each tool in Section B in S1 Text. While this figure only contains the most performant methods, a full list of methods investigated can be found in Section B in S1 Text, with their results described in Figs A and B in S1 Text. (f-j) Benchmarks of stochastic chemical kinetics SSA simulations of the five models. Via JumpProcesses.jl, Catalyst can use several different algorithms (e.g. Direct, Sorting Direct, RSSA, and RSSACR above) for exact Gillespie simulations. Here, the simulation runtime is plotted against the (physical) final time of the simulation. Due to their long runtimes, some tools were not benchmarked for the largest models. We note that, in [52], it was remarked that BioNetGen (dashed orange lines) use a pseudo-random number generator in SSAs that, while fast, is of lower quality than many (slower) modern generators such as Mersenne Twister. For full details on benchmarks, see Section 4.1.CC BY · PubMed Central
Fig 4 The simulation pipeline of a Catalyst model, with internal intermediates displayed. Code as written by the user (yellow background), and as generated internally by Catalyst and ModelingToolkit (blue and grey backgrounds respectively) are shown, in addition to the generated structures and their fields (blue background, some of the internal fields are omitted in all displayed structures). (a) A symbolic ReactionSystem for a reversible dimerisation reaction is created using either the DSL, or programmatically using the Symbolics computer algebraic system. (b) The ReactionSystem can be converted into a ModelingToolkit ODESystem structure, corresponding to a symbolic RRE ODE model. (c) By providing initial conditions, parameter values, and a time span, the ODESystem can be simulated, generating an output solution. The generated (internal) Julia code for evaluating the derivatives defining the ODEs, which gets compiled and is input to the ODE solver, is displayed in grey. At each step, the user has the ability to investigate and manipulate the generated structures.
Fig 4 The simulation pipeline of a Catalyst model, with internal intermediates displayed. Code as written by the user (yellow background), and as generated internally by Catalyst and ModelingToolkit (blue and grey backgrounds respectively) are shown, in addition to the generated structures and their fields (blue background, some of the internal fields are omitted in all displayed structures). (a) A symbolic ReactionSystem for a reversible dimerisation reaction is created using either the DSL, or programmatically using the Symbolics computer algebraic system. (b) The ReactionSystem can be converted into a ModelingToolkit ODESystem structure, corresponding to a symbolic RRE ODE model. (c) By providing initial conditions, parameter values, and a time span, the ODESystem can be simulated, generating an output solution. The generated (internal) Julia code for evaluating the derivatives defining the ODEs, which gets compiled and is input to the ODE solver, is displayed in grey. At each step, the user has the ability to investigate and manipulate the generated structures.CC BY · PubMed Central
Fig 5 A wide range of features are available for Catalyst model analysis. A CRN model can be created either through the DSL, by manually declaring the reaction events, or by loading it from a file. The model is stored in the ReactionSystem IR, which can be used as input to a wide range of methods. Purple boxes indicate code written by the user, and green boxes the corresponding output. For some methods, either one, or both, boxes are omitted.
Fig 5 A wide range of features are available for Catalyst model analysis. A CRN model can be created either through the DSL, by manually declaring the reaction events, or by loading it from a file. The model is stored in the ReactionSystem IR, which can be used as input to a wide range of methods. Purple boxes indicate code written by the user, and green boxes the corresponding output. For some methods, either one, or both, boxes are omitted.CC BY · PubMed Central

Verification

Claims the paper makes

6 Valid · 5 Not valid · 9 Not checked · 69 Own results and interpretations

37 more claims · 2 with their cited sources checked

Create a free account to see all 89 claims and to verify the 9 references still open.

See all claims — free account
  • Probably not a substantial claim. 34
  • It may state two separate claims as one. 19
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Citations

Cites

Show all 80 references

Research works printed but not linked: 4

  1. H Resat, 2009, Computational Systems Biology. Methods in Molecular Biology, 311
  2. P Breiding, 2018, International Congress on Mathematical Software, 458
  3. Veltz R. BifurcationKit.jl; 2020. https://hal.archives-ouvertes.fr/hal-02902346.
  4. Chen J, Revels J. Robust benchmarking in noisy environments. arXiv e-prints. 2016;.

Cited by

See all 47

Metadata

Authors
Loman, T. E., Ma, Y., Ilin, V., Gowda, S., Korsbo, N., Yewale, N., Rackauckas, C., & Isaacson, S. A.
Affiliations
  • University of Cambridge (GB)
  • University of Washington (US)
  • Massachusetts Institute of Technology (US)
  • Indian Institute of Technology Madras (IN)
  • Boston University (US)
Corresponding author
Isaacson, Samuel A.
Venue
PLoS Computational Biology
Published
2023-10-18
Volume
19
Issue
10
Pages
e1011530
ISSN
  • 1553-734X (ISSN-L)
  • 1553-7358
OpenAlex
W4387736994
Cited by
47
References
72
Citations by year
2026: 12 · 2025: 16 · 2024: 13 · 2023: 5
FWCI
4.08
Type
journal article
Language
en
Open access
Locations
  • PLoS Computational Biology (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • OpenBU (Boston University) (publishedVersion, cc-by)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion)
Funders
  • HORIZON EUROPE Marie Sklodowska-Curie Actions (721456)
  • Welcome Trust (223770/Z/21/Z)
  • Chan Zuckerberg Initiative (2021- 237457)
  • NSF-DMS (1902854)
  • 2021 Google Summer of Code Fellowship
  • National Science Foundation (OAC-1835443, ECCS-2029670, OAC-2103804, PHY-2021825, S002283-USAID)
  • Defense Advanced Research Projects Agency (DE-AR0001211, HR00112290091)
  • Advanced Research Projects Agency-Energy (DE-AR0001222)
  • U.S. Agency for International Development through Penn State (S002283-USAID)
  • The Research Council of Norway and Equinor AS (308817)
  • United States Air Force Research Laboratory and the United States Air Force Artificial Intelligence Accelerator (FA8750-19-2-1000)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
  • Physical Sciences › Mathematics › Modeling and Simulation › Modeling and Simulation Systems
  • Physical Sciences › Mathematics › Modeling and Simulation › Mathematical and Computational Methods
Keywords
chemical reaction networks, Julia, stochastic chemical kinetics, chemical Langevin equation, reaction rate equation, domain-specific languages, intermediate representation, symbolic modeling, code generation, high-performance simulation
MeSH terms
Algorithms, Computer Simulation, Models, Biological, Models, Theoretical, Stochastic Processes

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Catalyst: Fast and flexible modeling of reaction networks. https://app.cochat.ai/papers/10.1371/journal.pcbi.1011530
The paper
Loman, T. E., Ma, Y., Ilin, V., Gowda, S., Korsbo, N., Yewale, N., Rackauckas, C., & Isaacson, S. A. (2023). Catalyst: Fast and flexible modeling of reaction networks. PLoS Computational Biology. https://doi.org/10.1371/journal.pcbi.1011530

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.