Pith. sign in

REVIEW 3 major objections 4 minor 41 references

Simulating is not always understanding: When model complexity obscures biology

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A cell model generates understanding only when its free parameters are pinned down by experimental constraints and one can see why it behaves as it does; reproducing an observation alone is not enough.

desk verdict A well-argued perspective with a concrete illustration; the core message holds, but the extrapolation from a small oscillator module to whole-cell models is asserted more than shown. read the letter →

arxiv 2608.06998 v1 pith:GLOCE3H2 submitted 2026-08-07 q-bio.MN physics.bio-ph

Lendert Gelens , Alejandro Fábregas-Tejeda , Grant Ramsey , Sylvia Wenmackers , Bart Smeets This is my paper · ORCID
classification q-bio.MNphysics.bio-ph
keywords cellularmodelingparameteridentifiabilitysloppinesswhole-cellmodelsmodelhierarchiesbifurcationanalysiscellcycleoscillatorscientificunderstanding
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

A computational model earns scientific understanding only when it does something beyond reproducing the observations used to build it—predicting a new result, exposing an unsuspected coupling, or failing in a way that reveals a necessary component. The paper argues that the decisive feature is not model size or component count but identifiability: free parameters must be few enough relative to experimental constraints to be pinned down, and the relation between parameters and behavior must be graspable. A concrete comparison of two cell-cycle oscillator models shows a detailed 13-parameter version consistent with the observed period over about 85% of the scanned parameter space, while a minimal 3-parameter version built on a measured bistable response is consistent with only about 2%. On this view, whole-cell simulations are substantial technical achievements and valuable hypothesis generators, but they have not yet delivered mechanistic explanation because their parameter spaces have not been probed and reduced.

What carries the argument

The load-bearing object is parameter sloppiness, the near-universal phenomenon in systems-biology models in which a small number of stiff parameter combinations control predictions while most parameters can vary by orders of magnitude without affecting observable outputs. The paper's concrete demonstration is the comparison between the mass-action cell-cycle model and its minimal functional-response counterpart: in the detailed model the bistable switch is an emergent many-to-one function of 11 internal kinetic parameters, so fitting the oscillation period leaves the rates almost entirely unconstrained; in the minimal model the same switch enters as a fixed, directly measured input-output curve, so the remaining 3 parameters are informative. Sloppiness is what does the work of the argument: it explains why prediction can be robust while causal knowledge stays out of reach, and why adding mechanistic detail without adding experimental constraints makes a model harder, not easier, to learn from.

What would settle it

Run a systematic parameter scan or profile-likelihood analysis on a publicly available whole-cell model such as JCVI-syn3A, fixing an experimentally measured output like division time. If the model matches the output across most of its parameter space, the paper's claim is supported; if only a small fraction is consistent, its key assumption that sloppiness scales with model size fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that reproducing a known observation establishes consistency but not mechanism, and that a model becomes explanatory only when it makes a novel prediction, uncovers an unexpected coupling, or fails in a way that identifies a missing component. What determines whether a model can do any of these is not how many molecules or spatial dimensions it includes but the ratio of free parameters to independent experimental constraints and whether one can see why the model produces its behavior. The authors demonstrate the point with two models of the Xenopus embryonic cell cycle oscillator: a mass-action model with 5 variables and 13 free parameters reproduces the roughly 40-minute period across about 85% of the scanned synthesis-degradation parameter space because its 11 internal rates can be adjusted in flat, compensating directions, whereas a minimal model with 2 variables and 3 free parameters—built by fixing the measured bistable APC/C response—is consistent with only about 2% of the same space. From this they conclude that whole-cell models of Mycoplasma genitalium and JCVI-syn3A, however comprehensive, currently offer little mechanistic understanding; what they provide instead is higher-order evidence: an assembled, mutually consistent inventory of empirical knowledge that can guide hypothesis generation but not yet causal explanation.

Load-bearing premise

The critique of whole-cell models assumes that parameter sloppiness becomes at least as severe in models thousands of times larger as it is in the small two-variable cell-cycle module analyzed, so that identifiability does not improve with scale.

Editorial extensions

If this is right

  • Large agent-based models of cytoskeletal dynamics and vertex models of tissue mechanics can remain genuinely explanatory at scale, provided a small number of physically grounded parameters keep the parameter space explorable.
  • Whole-cell simulations should be treated as structured inventory-and-hypothesis tools rather than end-point explanations until they are subjected to systematic reduction and dynamical analysis.
  • A mechanistic claim from any model with many unconstrained parameters requires comparison with simpler representations; a good fit alone does not identify the mechanism.
  • Machine-learned and neural-differential-equation models inherit the same identifiability constraint: increasing expressive capacity without increasing experimental constraints worsens the parameter-data imbalance.
  • The two analytical practices—building explicit model hierarchies and mapping control parameters, bifurcations, and robustness boundaries—are the route that converts simulation into understanding.

Reading between the lines

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

  • A testable extension is that every mechanistic model should report an effective parameter dimension or identifiability measure alongside goodness of fit, allowing progress to be judged by how tightly parameters are constrained rather than by how many components are included.
  • If the argument generalizes, systematically reducing an existing whole-cell model should reveal that a small number of stiff parameter combinations suffice to reproduce any given phenotype, with most parameters free to float without changing the output.
  • The same logic predicts that 'virtual cell' foundation models will support causal intervention only if their training enforces identifiability constraints; interpolation accuracy alone will not ground mechanistic inference.
  • One could push the paper's logic further and propose a minimal whole-cell model in which each module's detailed kinetics is replaced by an experimentally measured input-output response function, exactly as the cell-cycle example replaces the PP2A-ENSA-GWL subnetwork.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. This perspective argues that model complexity in cell biology can obscure rather than enable understanding. The authors' central claim is that what matters for insight is not the number of components in a model but (1) the ratio of free parameters to available experimental constraints and (2) whether the relationship between parameters and behavior can be grasped. They illustrate this with a Xenopus cell-cycle oscillator: a 13-parameter mass-action model is consistent with the observed period across ~85% of a scanned parameter plane, while a minimal 3-parameter model built around a measured bistable functional response is consistent with only ~2% of the same plane. From this they conclude that whole-cell models, despite being impressive technical achievements and sources of higher-order evidence, have not yet delivered mechanistic understanding, and they propose explicit model hierarchies, dynamical analysis, and control-parameter identification as the route to such understanding. The argument is explicitly framed as contingent rather than principled, and the numerical code for Figure 2 is promised in a public repository.

Significance. If the argument holds, the paper would be a valuable corrective to the widespread assumption that adding biological detail automatically improves explanatory power. Its strengths are the clarity of the conceptual framework, the concrete and reproducible numerical illustration, the explicit acknowledgment that the whole-cell claim is contingent, and the meaningful engagement with the philosophical literature on minimal models and understanding. The paper also usefully distinguishes predictive success from causal understanding and connects this distinction to machine-learning models. As a perspective, its contribution is conceptual rather than technical, but the Figure 2 example and the proposed methodological agenda could influence modeling practice in systems and cell biology.

major comments (3)
  1. [The identifiability problem: a concrete illustration (Figure 2)] The 85% versus 2% acceptance fractions are point estimates from 5000 random draws, reported without confidence intervals or a sensitivity analysis over the scan ranges for the synthesis and degradation rates and over the 40±1 min acceptance window. As reported, the numbers depend on unspecified prior choices for the 11 internal parameters and on the random-draw protocol. The qualitative conclusion would survive, but the quantitative-looking claim should be accompanied by, at minimum, binomial confidence intervals and a brief statement of how the scan ranges were chosen.
  2. [Scale, parameters, and the conditions for insight; The identifiability problem: a concrete illustration] The extrapolation from a single 13-parameter embryonic oscillator module to whole-cell models is the load-bearing step for the paper's critique of whole-cell modeling, yet it rests on an appeal to intuition: 'If parameter sloppiness is already severe in a two-variable approximation of a single embryonic oscillator module, it is difficult to see how it could be less severe in a model that is larger by several orders of magnitude.' No quantitative scaling argument or identifiability measurement on an actual whole-cell model is provided. The explicit hedge that the claim is contingent softens this, but the conclusions still lean on the scaling sentence. The authors should either supply a quantitative analysis, draw more carefully on cited work such as Babtie and Stumpf (Ref. [20]) that addresses whole-cell parameter problems, or rephrase the whole-cell conclusion as an explicitly unverified conjecture about a class of models that has not yet been analyzed.
  3. [The identifiability problem: a concrete illustration, paragraph beginning 'It is worth being explicit about why this…] The rebuttal to the objection that the minimal model was simply given more information is not demonstrated. The authors assert that fitting the mass-action model to the measured bistability curve would merely select another flat, degenerate region of parameter space, but no simulation or calculation supports this claim. Because the fairness of the Figure 2 comparison is central to the identifiability argument, the rebuttal should be backed by a numerical experiment: for example, fit the mass-action model to the measured functional response and show that the posterior over the 11 internal parameters remains flat or that the period constraint still admits a large fraction of the parameter space. Without this, the example remains open to the simpler reading that the minimal model is more identifiable because it incorporates an additional experimental observable by construction.
minor comments (4)
  1. [Abstract] There are missing spaces in the abstract, for example 'handfulofparameters' and 'towhole-cellsimulations'; please correct these typographical errors throughout the manuscript.
  2. [Figure 1] The placement of 'Machine learning models' in the underconstrained region is not explained in the text. Since some machine-learning models are trained on very large datasets, the authors should add a sentence justifying this placement, for example by noting that the relevant constraint-to-parameter ratio refers to mechanistic parameters rather than training examples.
  3. [Scale, parameters, and the conditions for insight] The central term 'grasped' in the two-factor account of understanding is not defined. The later connection to interpretability helps, but a brief definition or pointer to the interpretability literature (Refs. [10-12]) would make the criterion more precise.
  4. [References [24,25] and Figure 2] The models and measurements used in Figure 2 come from the authors' own prior work (Refs. [24,25]). This is transparent through the citations, but the text should state it explicitly in the figure description so that readers can assess potential confirmation bias.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's Figure 2 illustration is a transparent, code-reproduced comparison, and the whole-cell extrapolation is an explicitly contingent argument by analogy, not a derived consequence.

full rationale

The paper is a perspective, not a derivation: it argues for a criterion (free parameters versus experimental constraints, plus interpretability) and applies it to model classes. The central illustration (Figure 2) compares a 13-parameter mass-action model and a 3-parameter phenomenological model of the same Xenopus oscillator. The authors explicitly state that the minimal model's measured functional response is 'fixed directly from bistability experiments [25]' and enters 'as a fixed, built-in element'; they then compute how much of the scanned (b_syn, b_deg) plane is consistent with the period constraint. This is an actual simulation result made available in a GitLab repository, not a fitted parameter renamed as a prediction. The asymmetry between the two models is defended, not hidden: the paper argues that the mass-action model cannot 'cleanly absorb' the bistability constraint because that observable is an emergent function of its parameters. Whether that argument is convincing is a scientific question, but it is not circular. The extrapolation to whole-cell models is the weakest point, but it is explicitly framed as contingent ('We stress that our claim is contingent, not principled') and as an appeal to intuition ('it is difficult to see how'), which is a gap in evidence, not a circular reduction. Self-citations to Refs. [24], [25], [12], and [41] support the illustration and data availability, but they are not load-bearing uniqueness theorems or unverified ansatze; the cited models are reproduced and the experimental curve is external. No equation in the paper reduces to another by construction, and no claimed prediction is identical to an input constraint.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The ledger captures the assumptions the perspective imports from prior literature. The paper does not fit a new theory to data; its free parameters are the tuning choices of the illustrative simulation in Figure 2. The axioms are the background results, empirical claims, and philosophical premises the argument depends on. No new entities are proposed.

free parameters (3)
  • Random-draw count for parameter scan = 5000
    The identifiability volumes (85% vs 2%) depend on this sampling effort; no convergence analysis is provided.
  • Period acceptance window = T = 40 +/- 1 min
    The acceptance criterion for the parameter scan; changing the window width would change the volume estimates.
  • Scan ranges for cyclin synthesis and degradation rates (b_syn, b_deg) = unspecified
    The boundaries of the scanned parameter space are not given in the text; they determine the 85% and 2% fractions and are chosen by the authors, presumably from Ref. [24].
assumptions (5)
  • domain assumption Understanding requires identifying causally responsible components and generalizing under counterfactual interventions (mechanistic explanation paradigm)
    Invoked in the opening section via citations [1-3]; the entire argument presupposes this notion of understanding.
  • domain assumption Parameter sloppiness is near-universal in systems biology models
    Taken from Refs [8,9,20] and used to infer that detailed models have large degenerate regions.
  • domain assumption The bistability measurement [25] fixes the minimal model's functional response independently of the oscillation period
    This is the load-bearing empirical input for the Figure 2 comparison; if the measured curve were itself period-dependent, the asymmetry between models would weaken.
  • domain assumption Whole-cell models have far more free parameters than independent experimental constraints; agent-based cytoskeletal models have small parameter spaces
    Asserted in Figure 1 caption and text based on [6,7,14,20]; no quantitative parameter/constraint census is performed.
  • domain assumption Levins' trade-off: generality, realism, and precision cannot be maximized simultaneously
    Cited [13] to frame the realism-understanding tension.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Simulating is not always understanding: When model complexity obscures biology." pith.science (2026). https://pith.science/paper/GLOCE3H2

@misc{pith2026260806998,
  author       = {Pith},
  title        = {Pith review of: Simulating is not always understanding: When model complexity obscures biology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GLOCE3H2}},
  note         = {Machine review of arXiv:2608.06998}
}
read the original abstract

In cell biology, computational models of biological systems range from minimal representations with a handful of parameters to whole-cell simulations tracking thousands of molecular species across a complete cell cycle. While these models span a continuum of detail, increasing complexity changes what they capture and are able to explain, what they can predict, and how they can fall short. A model contributes to understanding only when it makes novel predictions, reveals an unexpected coupling between processes, or fails in a way that identifies missing parameters. We contend that what is important for understanding is not the number of components or spatial dimensions a model contains, but the ratio of free parameters to the experimental constraints available to pin them down, and whether we can see why it produces the behaviors it does. Large-scale agent-based models of cytoskeletal dynamics or tissue mechanics that are built on a small number of physically grounded rules can reveal rich self-organization behavior precisely because their parameter spaces are small enough to explore systematically. By contrast, when free parameters grow faster than the data available to constrain them, models become progressively harder to interpret -- and even disprove --regardless of their biological scope. We argue that the field needs to reconsider the goal of complex models. We should move away from trying to include as many parameters as possible and instead aim for systematic comparisons with simpler representations, dynamical analysis, and explicit model hierarchies that trace how cellular behavior emerges from its parts.

Figures

Figures reproduced from arXiv: 2608.06998 by the authors.

Figure 1
Figure 1. Schematic representation of various model types in terms of model complexity and iden￾tifiability. These are conceptual maps, not based on precise measurements. On the left, model types are represented in terms of component count and parameter count, interpretability decreases as the number of parameters grows. On the right, two of the same model types reappear, one in which the free parameters are well-constrained … view at source ↗
Figure 2
Figure 2. A detailed and a minimal model of the cell cycle oscillator produce equivalent dynamics but differ strongly in parameter identifiability. (A) Mass-action model with 5 variables and 13 parameters. The network (left) captures CycB-Cdk1 driving its own destruction through a double-negative feedback loop involving GWL kinase, ENSA, and PP2A phosphatase. Simulations produce oscillations with period ∼40 min (middle, top).… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages

  1. [20]

    Babtie and Michael P

    Ann C. Babtie and Michael P. H. Stumpf. How to deal with parameters for whole-cell modelling. Journal of The Royal Society Interface, 14(133):20170237, 2017

    work page 2017

  2. [1]

    Dynamic mechanistic explanation: computational modeling of circadian rhythms as an exemplar for cognitive science.Studies in History and Philosophy of Science, 41:321–333, 2010

    William Bechtel and Adele Abrahamsen. Dynamic mechanistic explanation: computational modeling of circadian rhythms as an exemplar for cognitive science.Studies in History and Philosophy of Science, 41:321–333, 2010

    work page 2010

  3. [2]

    Carl F. Craver. When mechanistic models explain.Synthese, 153:355–376, 2006

    work page 2006

  4. [3]

    Cambridge University Press, Cambridge, 2nd edition, 2009

    Judea Pearl.Causality: Models, Reasoning and Inference. Cambridge University Press, Cambridge, 2nd edition, 2009

    work page 2009

  5. [4]

    Three kinds of idealization.Journal of Philosophy, 104:639–659, 2007

    Michael Weisberg. Three kinds of idealization.Journal of Philosophy, 104:639–659, 2007

    work page 2007

  6. [5]

    Modelling and representing: an artefactual approach to model-based representation

    Tarja Knuuttila. Modelling and representing: an artefactual approach to model-based representation. Studies in History and Philosophy of Science, 42:262–271, 2011

    work page 2011

  7. [6]

    Karr, Jayodita C

    Jonathan R. Karr, Jayodita C. Sanghvi, Derek N. Macklin, Miriam V. Gutschow, Jared M. Jacobs, Ben- jamin Bolival, Nacyra Assad-Garcia, John I. Glass, and Markus W. Covert. A whole-cell computational model predicts phenotype from genotype.Cell, 150:389–401, 2012

    work page 2012

  8. [7]

    Thornburg, Ansel Maytin, Jonghan Kwon, Troy A

    Zane R. Thornburg, Ansel Maytin, Jonghan Kwon, Troy A. Brier, Benjamin R. Gilbert, Enguang Fu, Yuqing L. Gao, James Quenneville, Tejal Wu, Handuo Li, et al. Bringing the genetically minimal cell to life on a computer in 4d.Cell, 189:2582–2597, 2026

    work page 2026

Show all 41 references
  1. [8]

    Gutenkunst, Joshua J

    Ryan N. Gutenkunst, Joshua J. Waterfall, Fergal P. Casey, Kevin S. Brown, Christopher R. Myers, and James P. Sethna. Universally sloppy parameter sensitivities in systems biology models.PLoS Computational Biology, 3:e189, 2007

    2007

  2. [9]

    Kamil Erguler and Michael P. H. Stumpf. Practical limits for reverse engineering of dynamical systems: a statistical analysis of sensitivity and parameter inferability in systems biology models.Molecular BioSystems, 7(5):1593–1602, 2011

    2011

  3. [10]

    Adynamicalparadigmformolecularcellbiology.TrendsinCellBiology, 30:504–515, 2020

    JohnJ.TysonandBelaNovak. Adynamicalparadigmformolecularcellbiology.TrendsinCellBiology, 30:504–515, 2020. 14

    2020

  4. [11]

    Understanding from machine learning models.British Journal for the Philosophy of Science, 73(1):109–133, 2022

    Emily Sullivan. Understanding from machine learning models.British Journal for the Philosophy of Science, 73(1):109–133, 2022

    2022

  5. [12]

    Data-driven discovery of dynamical models in biology.Nature Reviews Physics, 2026

    Bartosz Prokop and Lendert Gelens. Data-driven discovery of dynamical models in biology.Nature Reviews Physics, 2026

    2026

  6. [13]

    The strategy of model building in population biology.American Scientist, 54(4): 421–431, 1966

    Richard Levins. The strategy of model building in population biology.American Scientist, 54(4): 421–431, 1966

    1966

  7. [14]

    Nédélec, Thomas Surrey, Anthony C

    François J. Nédélec, Thomas Surrey, Anthony C. Maggs, and Stanislas Leibler. Self-organization of microtubules and motors.Nature, 389:305–308, 1997

    1997

  8. [15]

    On the explanatory power of atomistic simulations.European Journal for Philosophy of Science, 15:45, 2025

    Julie Schweer and Marcus Elstner. On the explanatory power of atomistic simulations.European Journal for Philosophy of Science, 15:45, 2025

    2025

  9. [16]

    Howtoinferexplanationsfromcomputersimulations.StudiesinHistoryandPhilosophy of Science Part A, 82:25–33, 2020

    FlorianJ.Boge. Howtoinferexplanationsfromcomputersimulations.StudiesinHistoryandPhilosophy of Science Part A, 82:25–33, 2020

    2020

  10. [17]

    Vertex models: from cell mechanics to tissue morphogenesis.Philosophical Transactions of the Royal Society B: Biological Sciences, 372(1720): 20150520, 2017

    Silvanus Alt, Poulami Ganguly, and Guillaume Salbreux. Vertex models: from cell mechanics to tissue morphogenesis.Philosophical Transactions of the Royal Society B: Biological Sciences, 372(1720): 20150520, 2017

    2017

  11. [18]

    Cristina Marchetti, Jean-François Joanny, Sriram Ramaswamy, Tanniemola B

    M. Cristina Marchetti, Jean-François Joanny, Sriram Ramaswamy, Tanniemola B. Liverpool, Jacques Prost, Madan Rao, and R. Aditi Simha. Hydrodynamics of soft active matter.Reviews of Modern Physics, 85:1143, 2013

    2013

  12. [19]

    Wendy S. Parker. Evidence and knowledge from computer simulation.Erkenntnis, 87:1521–1538, 2022

    2022

  13. [21]

    Wendy S. Parker. Model evaluation: an adequacy-for-purpose view.Philosophy of Science, 87(3): 457–477, 2020

    2020

  14. [22]

    The evolution of systems biology and systems medicine: from mechanistic models to uncertainty quantification.Annual Review of Biomedical Engineering, 27:425–447, 2025

    Lu Qiao et al. The evolution of systems biology and systems medicine: from mechanistic models to uncertainty quantification.Annual Review of Biomedical Engineering, 27:425–447, 2025

    2025

  15. [23]

    Banga and Alejandro F

    Julio R. Banga and Alejandro F. Villaverde. Mechanistic dynamic modelling of biological systems: the road ahead.Current Opinion in Systems Biology, 42:100553, 2025. 15

    2025

  16. [24]

    A modular approach for modeling the cell cycle based on functional response curves.PLoS Computational Biology, 17:e1009008, 2021

    Jolan De Boeck, Jan Rombouts, and Lendert Gelens. A modular approach for modeling the cell cycle based on functional response curves.PLoS Computational Biology, 17:e1009008, 2021

    2021

  17. [25]

    Bistable,biphasicregulationofPP2A-B55accounts for the dynamics of mitotic substrate phosphorylation.Current Biology, 31(4):794–808.e6, 2021

    JuliaKamenz,LendertGelens,andJamesE.Ferrell. Bistable,biphasicregulationofPP2A-B55accounts for the dynamics of mitotic substrate phosphorylation.Current Biology, 31(4):794–808.e6, 2021

    2021

  18. [26]

    Simulation and the sense of understanding

    Jaakko Kuorikoski. Simulation and the sense of understanding. In Paul Humphreys and Cyrille Imbert, editors,Models, Simulations, and Representations, pages 168–187. Routledge, New York, 2013

    2013

  19. [27]

    No understanding without explanation.Studies in History and Philosophy of Science, 44:510–515, 2013

    Michael Strevens. No understanding without explanation.Studies in History and Philosophy of Science, 44:510–515, 2013

    2013

  20. [28]

    de Regt.Understanding Scientific Understanding

    Henk W. de Regt.Understanding Scientific Understanding. Oxford University Press, New York, 2017

    2017

  21. [29]

    Batterman and Collin C

    Robert W. Batterman and Collin C. Rice. Minimal model explanations.Philosophy of Science, 81(3): 349–376, 2014

    2014

  22. [30]

    Complex versus complicated systems biology, universality versus detailed modeling

    Kunihiko Kaneko. Complex versus complicated systems biology, universality versus detailed modeling. Current Opinion in Systems Biology, 44:100589, 2026

    2026

  23. [31]

    Craver and David M

    Carl F. Craver and David M. Kaplan. Are more details better? on the norms of completeness for mechanistic explanations.British Journal for the Philosophy of Science, 71(1):287–319, 2020

    2020

  24. [32]

    Beyond reduction and pluralism: toward an epistemology of explanatory integration in biology.Erkenntnis, 73:295–311, 2010

    Ingo Brigandt. Beyond reduction and pluralism: toward an epistemology of explanatory integration in biology.Erkenntnis, 73:295–311, 2010

    2010

  25. [33]

    Theinfluence of cell mechanics, cell–cell interactions, and proliferation on epithelial packing.Current Biology, 17: 2095–2104, 2007

    RezaFarhadifar,Jens-ChristianRoper,BenoitAigouy,SuzanneEaton,andFrankJülicher. Theinfluence of cell mechanics, cell–cell interactions, and proliferation on epithelial packing.Current Biology, 17: 2095–2104, 2007

    2007

  26. [34]

    Burkhardt, et al

    Charlotte Bunne, Yusuf Roohani, Yanay Rosen, Ankit Gupta, Xikun Zhang, Marcel Roed, Theodoros Alexandrov, Mohammed AlQuraishi, Patrick Brennan, Daniel B. Burkhardt, et al. How to build the virtual cell with artificial intelligence: priorities and opportunities.Cell, 187:7045–7...

    2024

  27. [35]

    scgpt: toward building a foundation model for single-cell multi-omics using generative AI.Nature Methods, 21:1470–1480, 2024

    Haotian Cui, Chloe Wang, Hassaan Maan, Kuan Pang, Fengning Luo, Nan Duan, and Bo Wang. scgpt: toward building a foundation model for single-cell multi-omics using generative AI.Nature Methods, 21:1470–1480, 2024. 16

    2024

  28. [36]

    Brydon, Zexian Zeng, X

    ChristinaV.Theodoris,LingXiao,AnantChopra,MarkD.Chaffin,ZeinaR.AlSayed,MatthewC.Hill, Helene Mantineo, Elizabeth M. Brydon, Zexian Zeng, X. Shirley Liu, and Patrick T. Ellinor. Transfer learning enables predictions in network biology.Nature, 618:616–624, 2023

    2023

  29. [37]

    Towards the construction of a virtual yeast.Nature, 655: 59–70, 2026

    Liujia Qian, Zizhuo Zhou, Peijie Zhou, et al. Towards the construction of a virtual yeast.Nature, 655: 59–70, 2026

    2026

  30. [38]

    Garcia Gomez, Kirsten H

    Ben Noordijk, Monica L. Garcia Gomez, Kirsten H. ten Tusscher, Dick de Ridder, Aalt D. J. van Dijk, and Robert W. Smith. The rise of scientific machine learning: a perspective on combining mechanistic modelling with machine learning for systems biology.Frontiers in Systems Bio...

    2024

  31. [39]

    Artificial intelligence-enabled multi-scale virtual cell: perspective, challenges, and opportunities.Briefings in Bioinformatics, 27(bbag104), 2026

    Huasen Jiang, Xiaoyu Huang, Xiangpeng Bi, Wenjian Ma, Haibo Ni, Zhiqiang Wei, Pin Sun, Henggui Zhang, and Shugang Zhang. Artificial intelligence-enabled multi-scale virtual cell: perspective, challenges, and opportunities.Briefings in Bioinformatics, 27(bbag104), 2026

    2026

  32. [40]

    Theis, and Bo Wang

    Haotian Cui, Alejandro Tejada-Lapuerta, Maria Brbić, Julio Saez-Rodriguez, Simona Cristea, Hani Goodarzi, Mohammad Lotfollahi, Fabian J. Theis, and Bo Wang. Towards multimodal foundation models in molecular cell biology.Nature, 640:623–633, 2025

    2025

  33. [41]

    Gitlab repository.https://gitlab.kuleuven.be/gelenslab/publications/ 2026_Gelens_perspective_simulating_and_understanding.git, 2026

    Lendert Gelens. Gitlab repository.https://gitlab.kuleuven.be/gelenslab/publications/ 2026_Gelens_perspective_simulating_and_understanding.git, 2026. 17

    2026

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.