pith. sign in

arxiv: 2506.19866 · v1 · submitted 2025-06-10 · 🧬 q-bio.MN · cs.PF· math.OC· q-bio.QM

GPU-accelerated Modeling of Biological Regulatory Networks

Joyce Reimer , Pranta Saha , Chris Chen , Neeraj Dhar , Brook Byrns , Steven Rayan , Gordon Broderick This is my paper

Pith reviewed 2026-05-19 09:46 UTC · model grok-4.3

classification 🧬 q-bio.MN cs.PFmath.OCq-bio.QM
keywords GPU accelerationbiological regulatory networkslogic modelsglobal optimizationparameter searchdiscrete dynamical systemsin silico modeling
0
0 comments X

The pith

GPU implementations of global optimization algorithms accelerate parameter searches for biological logic models by 33 to 1866 percent over CPU methods.

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

This paper shows that moving global optimization schemes to a GPU environment substantially speeds up the search for parameters in discrete logic models that fit sparse time-course data from biological regulatory networks. The authors apply the approach to two model systems that differ by nearly an order of magnitude in scale and measure clear efficiency gains against both serial and multi-threaded CPU baselines. If the gains hold at larger scales, the technique makes logic-model identification practical for in silico hypothesis generation and experiment design in pharmaceutical research rather than remaining limited by computation time.

Core claim

The implementation of global optimization algorithms in a GPU-computing environment accelerates the solution of parameter search problems for discrete logic models of biological regulatory networks, yielding 33%-43% improvement compared to multi-thread CPU implementations and 33%-1866% increase compared to CPU in serial on two model systems that represent almost an order of magnitude scale-up in complexity.

What carries the argument

GPU-accelerated global optimization applied to fitting discrete logic models to time-course data from regulatory networks

If this is right

  • Parameter searches become feasible for regulatory networks of greater size without requiring prohibitive runtimes.
  • Researchers can generate predictions about network behavior under varying conditions on shorter timescales.
  • In silico hypothesis generation and experiment design gain viability as routine tools in pharmaceutical research.
  • Model identification from sparse data can support iterative refinement of regulatory hypotheses within practical time limits.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same GPU porting strategy could be tested on continuous or hybrid models of regulatory dynamics to check for comparable gains.
  • Smaller labs without large CPU clusters might now run ensemble fits that quantify uncertainty in the recovered logic rules.
  • Integration with automated experiment design loops could become realistic if the reduced runtimes allow many candidate models to be evaluated per cycle.

Load-bearing premise

The two chosen model biological regulatory systems are representative of the scale and structure of networks that would be used in practical in silico pharmaceutical research so the observed speedups generalize without major algorithm changes.

What would settle it

Running identical parameter searches on a third regulatory network at least twice as large as the larger of the two tested models and checking whether the GPU-to-CPU runtime ratios remain inside the reported 33-1866 percent improvement band.

Figures

Figures reproduced from arXiv: 2506.19866 by Brook Byrns, Chris Chen, Gordon Broderick, Joyce Reimer, Neeraj Dhar, Pranta Saha, Steven Rayan.

Figure 1
Figure 1. Figure 1: Unparameterized structure of the HPG (a) and M. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Performance profile (a) and acceptance rate (b) of [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Performance profile (a) and acceptance rate (b) of [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Simulated time course data compared to reference data, derived from an optimized HPG-axis regulatory network model. One of the solutions proposed by the optimization with the best objective function evaluation is plotted here, this one from the 1024-thread GPU configuration. It results in an objective function value of 20 (weighted Manhattan distance) and an unweighted Manhattan distance of 14 from the exp… view at source ↗
Figure 5
Figure 5. Figure 5: Simulated time course data compared to reference data, derived from an optimized M. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
read the original abstract

The complex regulatory dynamics of a biological network can be succinctly captured using discrete logic models. Given even sparse time-course data from the system of interest, previous work has shown that global optimization schemes are suitable for proposing logic models that explain the data and make predictions about how the system will behave under varying conditions. Considering the large scale of the parameter search spaces associated with these regulatory systems, performance optimizations on the level of both hardware and software are necessary for making this a practical tool for in silico pharmaceutical research. We show here how the implementation of these global optimization algorithms in a GPU-computing environment can accelerate the solution of these parameter search problems considerably. We carry out parameter searches on two model biological regulatory systems that represent almost an order of magnitude scale-up in complexity, and we find the gains in efficiency from GPU to be a 33%-43% improvement compared to multi-thread CPU implementations and a 33%-1866% increase compared to CPU in serial. These improvements make global optimization of logic model identification a far more attractive and feasible method for in silico hypothesis generation and design of experiments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims that GPU implementations of global optimization algorithms for inferring discrete logic models of biological regulatory networks from time-course data yield substantial speedups, specifically 33%-43% over multi-threaded CPU baselines and 33%-1866% over serial CPU on two model systems representing nearly an order-of-magnitude increase in complexity; these gains are presented as making the approach practical for in silico pharmaceutical research.

Significance. If the reported speedups are substantiated with reproducible benchmarks, the work would be significant for systems biology by addressing the computational bottleneck of large parameter spaces in logic-based network modeling, potentially enabling wider adoption for hypothesis generation and experiment design. The emphasis on hardware-level optimizations for sparse regulatory systems is a timely contribution, though the absence of detailed empirical support limits immediate impact assessment.

major comments (2)
  1. [Abstract] Abstract: the concrete performance claims (33%-43% improvement over multi-thread CPU and up to 1866% over serial) are stated without any benchmark tables, error bars, implementation details, or description of how the multi-thread CPU baseline was configured (e.g., thread count, hardware specs, or optimization flags). This directly undermines support for the central empirical claim of considerable acceleration.
  2. [Results] The manuscript tests only two model systems and does not examine whether the observed GPU speedups generalize to larger node counts, denser interaction graphs, or more complex logic functions typical of pharmaceutical-scale networks; memory divergence or host-device transfer overheads could alter efficiency, but no scaling analysis or stress tests on such regimes are provided.
minor comments (2)
  1. [Methods] Notation for the logic functions and parameter search space could be clarified with an explicit example in the methods to aid readers unfamiliar with discrete regulatory models.
  2. [Abstract] The abstract mentions 'almost an order of magnitude scale-up in complexity' but does not quantify this (e.g., number of nodes, edges, or parameters) in a table or equation for easy comparison.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful review and constructive comments on our manuscript. We have addressed each major comment point by point below, providing clarifications and indicating revisions where the manuscript will be updated to strengthen the presentation of our results.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the concrete performance claims (33%-43% improvement over multi-thread CPU and up to 1866% over serial) are stated without any benchmark tables, error bars, implementation details, or description of how the multi-thread CPU baseline was configured (e.g., thread count, hardware specs, or optimization flags). This directly undermines support for the central empirical claim of considerable acceleration.

    Authors: We agree that the abstract would benefit from additional context to support the performance claims. The detailed benchmark tables, including timings with error bars, hardware specifications (NVIDIA GPU model and multi-core CPU with explicit thread counts), and optimization flags (e.g., -O3 for CPU and CUDA compilation settings), are presented in the Results section along with the experimental setup. To address this, we have revised the abstract to include a brief reference to the hardware configurations used for the multi-threaded CPU baseline and to direct readers to the full benchmark details and tables in the main text. revision: yes

  2. Referee: [Results] The manuscript tests only two model systems and does not examine whether the observed GPU speedups generalize to larger node counts, denser interaction graphs, or more complex logic functions typical of pharmaceutical-scale networks; memory divergence or host-device transfer overheads could alter efficiency, but no scaling analysis or stress tests on such regimes are provided.

    Authors: The two model systems were selected specifically because they represent nearly an order-of-magnitude increase in complexity over prior examples, providing a practical demonstration of the GPU approach on representative biological regulatory networks. We acknowledge that the current study does not include a comprehensive scaling analysis across larger node counts, denser graphs, or more complex logic functions, nor explicit stress tests for memory divergence and host-device transfer overheads. We have added a new paragraph in the Discussion section that explicitly addresses these potential limitations, discusses the relevance of such factors for pharmaceutical-scale applications, and outlines directions for future work on generalization and performance characterization in those regimes. revision: partial

Circularity Check

0 steps flagged

Empirical runtime benchmarks contain no circular derivation

full rationale

The paper reports measured wall-clock speedups (33-43% over multi-thread CPU, up to 1866% over serial) obtained by executing the same global optimization code on GPU versus CPU hardware for two concrete logic-model identification tasks. These figures are direct experimental observations, not quantities derived from equations, fitted parameters, or prior self-citations that would reduce to the inputs by construction. No self-definitional loops, uniqueness theorems, or ansatz smuggling appear in the derivation chain; the central performance claims remain externally falsifiable by re-running the reported benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The acceleration claim rests on the domain assumption that the global optimization workload is sufficiently parallelizable for GPU hardware and that the two test networks capture the relevant complexity scaling; no new free parameters or invented entities are introduced by the paper itself.

axioms (1)
  • domain assumption Global optimization schemes are suitable for proposing logic models that explain the data and make predictions
    Invoked in the opening paragraph as established by previous work.

pith-pipeline@v0.9.0 · 5744 in / 1182 out tokens · 35589 ms · 2026-05-19T09:46:40.929842+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    We show here how the implementation of these global optimization algorithms in a GPU-computing environment can accelerate the solution of these parameter search problems considerably... simulated annealing algorithm... objective function y = min[( |S(x) − R| ) ⊙ ( |G| + 1) ]

  • IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    discrete logic formulation... state values of nodes can take on one of three activation values... priority update scheme

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages

  1. [1]

    Gosztyla, Xinzhu Tan, et al

    Yuan Mei, Maya L. Gosztyla, Xinzhu Tan, et al. 2025. Integrated multi-omic characterizations of the synapse reveal RNA processing factors and ubiquitin ligases associated with neurodevelopmental disorders. Cell Syst 16, 4, Article 101204 (April 2025). http://dx.doi.org/10.1016/j.cels.2025.101204

    work page doi:10.1016/j.cels.2025.101204 2025

  2. [2]

    Serrano, Adrian Pelin, Tabiwang N

    Lia R. Serrano, Adrian Pelin, Tabiwang N. Arrey, et al. 2025. Affinity purification mass spectrometry on the Orbitrap-Astral mass spectrometer enables ACM BCB’25, October 2025, Philadelphia, Pennsylvania USA J. Reimer et al. high-throughput protein-protein interaction mapping. J. Proteome Res 24, 4 (March 2025), 2006-2016. https://doi.org/10.1021/acs.jpro...

    work page doi:10.1021/acs.jproteome.4c01040 2025

  3. [3]

    Alexiou Athanasios, Vairaktarakis Charalampos, Tsiamis Vasileios and Ghulam Md Ashraf. 2017. Protein-protein interaction (PPI) network: recent advances in drug discovery. Curr. Drug Metab 18, 1 (Jan. 2017), 5-10. https://doi.org/10.2174/138920021801170119204832

    work page doi:10.2174/138920021801170119204832 2017

  4. [4]

    Craddock, Darrell Whitley and Gordon Broderick

    Hooman Sedghamiz, Matthew Morris, Travis J.A. Craddock, Darrell Whitley and Gordon Broderick. 2019. Bio-ModelChecker: using bounded constraint satisfaction to seamlessly integrate observed behavior with prior knowledge of biological networks. Front. Bioeng. Biotechnol 7, 48 (March 2019). https://doi.org/10.3389/fbioe.2019.00048

    work page doi:10.3389/fbioe.2019.00048 2019

  5. [5]

    René Thomas and Richard d’Ari. 1990. Biological Feedback. CRC Press, Inc

    work page 1990

  6. [6]

    Monteiro, Julio Saez- Rodriguez, Tomá š Helikar, Denis Thieffry and Claudine Chaouiya

    Wassim Abou-Jaoudé, Pauline Traynard, Pedro T. Monteiro, Julio Saez- Rodriguez, Tomá š Helikar, Denis Thieffry and Claudine Chaouiya. 2016. Logical modeling and dynamical analysis of cellular networks. Front. Genet 7, 94 (May 2016). https://doi.org/10.3389/fgene.2016.00094

    work page doi:10.3389/fgene.2016.00094 2016

  7. [7]

    Jonas Béal, Lorenzo Pantolini, Vincent Noël, Emmanuel Barillot and Laurence Calzone. 2021. Personalized logical models to investigate cancer response to BRAF treatments in melanomas and colorectal cancers. Comput. Biol 17, 1, Article e1007900 (Jan. 2021). https://doi.org/10.1371/journal.pcbi.1007900

    work page doi:10.1371/journal.pcbi.1007900 2021

  8. [8]

    Anton Feenstra et al

    Nicola Bonzanni, Abhishek Garg, K. Anton Feenstra et al. 2013. Hard-wired heterogeneity in blood stem cells revealed using a dynamic regulatory network model. Bioinformatics 29, 13 (July 2013), i80-888. https://doi.org/10.1093/bioinformatics/btt243

    work page doi:10.1093/bioinformatics/btt243 2013

  9. [9]

    Lyman, Matthew M

    Cole A. Lyman, Matthew M. Morris, Spencer Richman, Hongbao Cao, Antony Scerri, Chris Cheadle and Gordon Broderick. 2021. High fidelity modeling of pulse dynamics using logic networks. In 2021 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), December 9-12, 2021, Houston, Texas. IEEE, 197-204. https://doi.org/10.1109/BIBM52615.2021.9669437

    work page doi:10.1109/bibm52615.2021.9669437 2021

  10. [10]

    Anton Dekkers and Emile Aarts. 1991. Global optimization and simulated annealing. Math. Progr 50 (March 1991), 367-393. https://doi.org/10.1007/BF01594945

    work page doi:10.1007/bf01594945 1991

  11. [11]

    Kirkpatrick, C

    S. Kirkpatrick, C.D. Gelatt, Jr. and M.P. Vecchi. 1983. Optimization by simulated annealing. Science 220, 4598 (May 1983), 671-680. https://doi.org/10.1126/science.220.4598.671

    work page doi:10.1126/science.220.4598.671 1983

  12. [12]

    Ferreiro, J.A

    A.M. Ferreiro, J.A. García, J.G. López-Salas and C. Vásquez. 2012. An efficient implementation of parallel simulated annealing algorithm in GPUs. J. Glob. Optim, 57 (Sept. 2012), 863-890. https://doi.org/10.1007/s10898-012-9979-z

    work page doi:10.1007/s10898-012-9979-z 2012

  13. [13]

    Tias Guns. 2019. Increasing modeling language convenience with a universal n- dimensional array, CPpy as python-embedded example. In Proceedings of the 18th workshop on Constraint Modelling and Reformulation at CP (ModRef 2019), September 30, 2019, Stamford, Connecticut

    work page 2019

  14. [14]

    Siu Kwan Lam, Antoine Pitrou and Stanley Seibert. 2015. Numba: a llvm-based python jit compiler. In Proceedings of the Second Workshop on the LLVM Compiler Infrastructure in HPC (LLVM ’15), November 15, 2015, Austin, Texas. ACM Inc., New York, NY, 1-6. https://doi.org/10.1145/2833157.2833162

    work page doi:10.1145/2833157.2833162 2015

  15. [15]

    Han, B., Li, Y ., & Li, G.,

    Robert I. McLachlan, Nancy L. Cohen, Kristine D. Dahl, William J. Bremner and Michael R. Soules. 1990. Serum inhibin levels during the periovulatory interval in normal women: relationships with sex steroid and gonadotrophin levels. Clin. Endocrinol (Oxf) , 32, 1 (Jan. 1990), 39-48. https://doi.org/10.1111/j.1365- 2265.1990.tb03748.x

    work page doi:10.1111/j.1365- 1990

  16. [16]

    Bappaditya Dey and William R. Bishai. 2014. Crosstalk between Mycobacterium tuberculosis and the host cell. Semin. Immunol, 26, 6 (Dec. 2014), 486-496. https://doi.org/10.1016/j.smim.2014.09.002

    work page doi:10.1016/j.smim.2014.09.002 2014

  17. [17]

    Bachman, Benjamin M

    John A. Bachman, Benjamin M. Gyori and Peter K. Sorger. 2023. Automated assembly of molecular mechanisms at scale from text mining and curated databases. Mol Syst. Biol , 19, Article e11325 (March 2023). https://doi.org/10.15252/msb.202211325

    work page doi:10.15252/msb.202211325 2023

  18. [18]

    Jiuhai Chen and Jonas Mueller. 2024. Quantifying uncertainty in answers from any language model and enhancing their trustworthiness. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1:Long Papers) , August 2024, Bangkok, Thailand. Association for Computational Linguistics, 5186-5200. https://doi.org/10.186...

    work page doi:10.18653/v1/2024.acl- 2024

  19. [19]

    Tair Askar, Bekdaulet Shukirgaliyev, Martin Lukac and Ernazar Abdikamalov

    work page

  20. [20]

    Computation, 9, 12, Article 142 (December 2021)

    Evaluation of pseudo-random number generation on GPU cards. Computation, 9, 12, Article 142 (December 2021). https://doi.org/10.3390/computation9120142

    work page doi:10.3390/computation9120142 2021

  21. [21]

    Manssen, M

    M. Manssen, M. Weigel and A.K. Hartmann. 2012. Random number generators for massively parallel simulations on GPU. Eur. Phys. J. Spec. Top, 210 (September 2012), 53-71. https://doi.org/10.1140/epjst/e2012-01637-8

    work page doi:10.1140/epjst/e2012-01637-8 2012

  22. [22]

    Emran Bajrami, Maida Ašić, Emir Cogo, Dino Trnka and Novica Nosovic. 2012. Performance comparison of simulated annealing algorithm execution on GPU and CPU. In 2012 Proceedings of the 35th International Convention MIPRO , May 21-25, 2012, Opatija, Croatia. IEEE, 1785-1788

    work page 2012

  23. [23]

    Davis, Erik Verschueren, Gwendolyn M

    Zoe H. Davis, Erik Verschueren, Gwendolyn M. Jang, et al. 2015. Global mapping of herpesvirus-host protein complexes reveals a transcription strategy for late genes. Mol. Cell , 57, 2 (Jan. 2015), 349-60. https://doi.org/10.1016/j.molcel.2014.11.026

    work page doi:10.1016/j.molcel.2014.11.026 2015

  24. [24]

    Penn, Zoe Netter, Jeffrey R

    Bennett H. Penn, Zoe Netter, Jeffrey R. Johnson et al. 2018. An Mtb-human protein-protein interaction map identifies a switch between host antiviral and antibacterial responses. Mol. Cell, 71, 4 (Aug. 2018), 637-648. https://doi.org/10.1016/j.molcel.2018.07.010

    work page doi:10.1016/j.molcel.2018.07.010 2018