Pith. sign in

REVIEW 3 major objections 3 minor 1 cited by

Collective gene dynamics leave signatures of decision landscapes in cell fate coordinates

T0 review · 3 major / 3 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Cell fate trajectories in single-cell data carry signatures of universal decision landscapes, with straight paths flagging double cusps, curved paths flagging heteroclinic flips, and mixed clusters flagging triple cusps.

desk verdict A genuinely new Hopfield–landscape construction whose experimental class signatures are interesting but not yet diagnostic. read the letter →

arxiv 2506.04219 v2 pith:32S76LNT submitted 2025-06-04 q-bio.CB cond-mat.dis-nncond-mat.stat-mech

Maria Yampolskaya , Laertis Ikonomou , Pankaj Mehta This is my paper
classification q-bio.CBcond-mat.dis-nncond-mat.stat-mech MSC 37N2592C42
keywords cellfatelandscapesHopfieldnetworksscTOPorderparameterssingle-cellRNA-seqdecision-makingclassesbifurcationtheoryWaddingtonlandscapehematopoiesis
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

Single-cell RNA-sequencing time series can be read as trajectories in a low-dimensional cell fate space, and this paper tries to show that the shapes of those trajectories reveal which universal decision-making class a cell fate transition belongs to. Using generalized Hopfield order parameters as coordinates, the authors simulate archetypal trajectories for three landscape classes and then match them to experimental data: straight basophil-to-neutrophil paths, curved monocyte-to-neutrophil paths, and a mixed AT1/AT2 cluster in the developing lung are consistent with a double cusp, a heteroclinic flip, and a triple cusp, respectively. If the correspondences hold, cell fate decisions can be classified directly from scRNA-seq time series, and the same landscape classes can explain spatial patterning through lateral inhibition signals.

What carries the argument

The load-bearing object is the generalized Hopfield order parameter $m^\mu = \sum_\nu (A^{-1})^{\mu\nu} m_\nu$, with $A_{\mu\nu} = \sum_i \xi_{\mu i} \xi_{\nu i}$, which projects a cell's gene expression vector onto the subspace spanned by reference cell type profiles and defines cell fate coordinates. The dynamics are $\tau \, dm^\mu/dt = \sigma^\mu(-\beta \, \partial V/\partial m^\mu) - m^\mu$, where $\sigma^\mu$ is a softmax nonlinearity and the potential $V$ combines an inverted parabola with a signal-dependent landscape $\tilde V$ built from normal forms of elementary bifurcations. The signaling parameters $f, k$ tilt the landscape and destabilize attractors, producing the bifurcations that move cells from one fate to another.

What would settle it

Re-analyze the lineage-traced hematopoiesis and lung data while tracking the perpendicular component $x_i^\perp(t) = x_i(t) - \sum_\mu m^\mu(t)\,\xi_{\mu i}$; if this component changes systematically during any of the three transitions, or if adding a few reference profiles orthogonal to the current basis turns a straight trajectory curved, the projection assumption fails and the class assignments are not trustworthy.

Watch

Extended reading notes

Core claim

The paper's central claim is that the measured cell fate dynamics are consistent with developmental landscapes containing intermediate progenitors and saddle points, and that the specific trajectory geometry encodes the class. In cell fate coordinates, straight trajectories to final fates indicate a double cusp with no mixed state; curved trajectories through multilineage states indicate a heteroclinic flip with an unstable manifold; and a dense intermediate cluster between fates indicates a triple cusp with a transiently stable progenitor. Applied to lineage-traced hematopoiesis and developing lung alveolar cells, the data show all three signatures. The same landscape logic is used to distinguish two Notch-dependent models of airway patterning, where a single signal corresponds to a double cusp and two sequential signals to a triple cusp.

Load-bearing premise

The analysis assumes that the only gene expression relevant to a fate decision is the part that lies in the subspace spanned by the chosen reference cell types; the perpendicular component is treated as static and ignored, so if genes or states outside that reference basis drive a transition, the observed trajectories would not reflect the true landscape.

Editorial extensions

If this is right

  • If the correspondences hold, scRNA-seq time series can be classified by trajectory shape alone, without fitting a full landscape or choosing marker genes by hand.
  • The triple-cusp assignment for alveolar maturation implies a transient AT1/AT2 progenitor that is stable before birth and destabilized when air breathing begins, a testable in vitro prediction.
  • In airway injury, the two Notch-signaling models predict different commitment timing: double cusp commits cells early, triple cusp keeps them in a mixed state until a second signal.
  • The method is designed to scale to atlas-level data, so the same signatures could be used to survey developmental transitions across organs and species.
  • Different classes predict different signal sensitivities: heteroclinic flips are the most sensitive to fate-biasing signals, whereas double cusps resist multilineage expression.

Reading between the lines

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

  • An editor's extension: the straight-versus-curved distinction may serve as a practical lineage-relationship diagnostic, with straight paths suggesting distantly related fates and curved paths suggesting adjacent fates connected by a saddle or progenitor.
  • The projection-based coordinates depend on the reference basis chosen; systematically perturbing the basis, for example by adding or removing closely related cell types, would quantify how robust each class assignment is.
  • Because the same potential can be tilted by different signaling schedules, controlled in vitro differentiation with measured time courses could be used to decide between competing landscape classes for the same pair of fates.
  • The framework could be extended to multi-fate decisions by building higher-dimensional normal forms, though the paper restricts itself to three-attractor classes.
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

3 major / 3 minor

Summary. The paper introduces a phenomenological framework that links high-dimensional gene expression dynamics to low-dimensional cell fate landscapes. It generalizes modern Hopfield networks by adding signal-dependent potentials built from normal forms of elementary bifurcations, and defines scTOP generalized order parameters as cell fate coordinates. The authors simulate three decision-making classes—double cusp, triple cusp, and heteroclinic flip—and propose qualitative trajectory signatures for each. They then apply the framework to two published scRNA-seq time-series datasets (hematopoietic lineage tracing and developing mouse lung) and to a spatial patterning model of Notch-mediated airway differentiation, concluding that the measured cell fate dynamics are consistent with landscapes containing intermediate progenitors and saddle points.

Significance. If the proposed mapping from landscape class to trajectory shape were quantitatively validated, the framework would provide a useful bridge between bifurcation-theoretic decision classes and transcriptome-wide single-cell data, potentially enabling universal comparisons across cell fate transitions. The mathematical derivation in Eqs. 1–12 is internally consistent, the model is self-contained and uses externally defined scTOP coordinates without fitting model parameters to the experimental data, and the code and data availability statements are concrete. The spatial patterning prediction in Section II D is a falsifiable extension. However, the significance is currently limited because the central experimental inferences rest on visual pattern matching rather than quantitative tests, and the manuscript itself acknowledges that different landscapes can produce identical trajectories in certain regimes.

major comments (3)
  1. [II C and SI C] The central inference from experimental cell fate trajectories to specific decision-making classes is not quantitatively established. The assignments in Section II C (basophil-neutrophil as double cusp, monocyte-neutrophil as heteroclinic flip, AT1/AT2 as triple cusp) are based on visual inspection of scatter plots without error bars, null models, or alternative-class simulations. Since SI C states that 'different landscapes can produce the exact same trajectories in certain regimes' and that it is easier to eliminate classes than to identify them, the qualitative signatures described in Figure 3 do not by themselves select a class. A quantitative comparison—for example, computing summary statistics such as path curvature, intermediate-region density, or time spent in intermediate states for simulated ensembles of each class and comparing them with the experimental m^mu trajectories—is needed to support the class assignments.
  2. [II C 2 and SI B 1] The triple-cusp interpretation of the developing lung data relies on the presence of an intermediate AT1/AT2 cluster, but the authors themselves note that this cluster could be a saddle point rather than a transient attractor. The claim that it is a transiently stable mixed-state progenitor is supported only by the hand-picked simulation schedules: SI B1 states that the prominence of the intermediate cluster depends on inserting a 200-step gap between the two signals and that shortening the gap makes the intermediate state difficult to detect. No statistical test is provided to show the cluster is significantly more populated than expected under a double-cusp or heteroclinic-flip model with a slowly passing trajectory. The 'transient AT1/AT2 mixed-state progenitor' is therefore an invented entity whose existence is not demonstrated by the data presented.
  3. [II A, Eq. (4)] The model's projection onto the p-dimensional cell fate subspace is load-bearing for all three experimental conclusions, yet the manuscript does not validate that the perpendicular component x_perp is dynamically irrelevant. Equations (5)–(7) show that x_perp is static under the model dynamics, but this is a modeling assumption, not an empirical fact. If genes or regulatory programs outside the scTOP reference basis contribute to fate transitions, the observed m^mu trajectories would not faithfully reflect the true landscape. The paper should provide evidence that conclusions are robust to the choice of reference basis, for example by repeating the analysis with shuffled or augmented reference profiles and showing that the qualitative signatures persist.
minor comments (3)
  1. [Throughout] Several cross-references are unresolved placeholders, including 'SI section??' in Section II B, 'Figure??' in the caption of Figure 4, and 'section??' in Section II A. These should be replaced with actual references before publication.
  2. [II A, Eqs. (8)–(12)] The notation for the potential is confusing: V in Eq. (8) is the inverted parabola, while V with a tilde in the following paragraph is the signal-dependent term and the full potential in Eqs. (10)–(12) appears to include both. Defining the full potential explicitly and distinguishing it from the bare Hopfield potential would improve readability.
  3. [II C 1] The analysis of hematopoietic data is restricted to clonal families with exactly two final fates, but the text does not state how many clones met this criterion or how the 'effectively bipotent' families were selected from the lineage-tracing data. Reporting the number of clones and the selection criteria would aid reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the model is a forward simulation over externally specified normal forms, and experimental comparisons are qualitative and explicitly non-identifying.

full rationale

The derivation chain runs from externally established elementary catastrophe normal forms (Rand et al. [23], Saez et al. [21]) through the paper's own Hopfield-style update equations (Eqs. 5-12), which are derived rather than fitted. The scTOP coordinates are defined in Methods A with the projection formula x_i = sum_mu m^mu xi_mu i + x_perp_i; the citation to Yampolskaya et al. [29] is corroborated by these equations and by the publicly released code, so it does not function as an unverified load-bearing premise. No experimental trajectory is used to fit landscape coefficients or signaling schedules; the schedules in Methods B and SI B1 are hand-specified to illustrate class features. The paper repeatedly acknowledges that class topology does not determine trajectories and that different landscapes can give the same trajectories (SI C), so the experimental section is framed as consistency rather than identification. Thus no step reduces by construction to its inputs.

Assumptions & free parameters 6 free parameters · 6 assumptions · 1 invented entities

The central claim rests on the scTOP projection being a faithful cell fate coordinate system, the softmax Hopfield dynamics being a valid map from landscapes to trajectories, and a set of manually chosen landscape coefficients and signal schedules. No parameters are fitted to the experimental data, which keeps the circularity burden low, but the manual choices reduce the strength of the signature-to-class inference.

free parameters (6)
  • Signaling parameter f (bias toward final fates) = Double cusp: 0 to ±0.1; triple cusp: 0 to ±0.3; heteroclinic flip: 0 to ±0.5
    Manually ramped over simulation windows to tilt the landscape; not fitted to data. The sign is chosen randomly per trial.
  • Signaling parameter k (destabilizes initial attractor) = Double cusp: 0.15; triple cusp: 0 to 0.3; heteroclinic flip: 0.5 to 2
    Manually set to trigger bifurcations; controls time spent at intermediate or saddle states.
  • Signal timing windows (start and end times for f and k) = Triple cusp: k over steps 0-400, f over 600-700; double cusp: f over 0-750; heteroclinic flip: k 100-500, f 450-600
    Chosen to emphasize class-defining features; the gap between signals controls whether the intermediate cluster is detectable.
  • Landscape polynomial coefficients = V_double=x^4+y^4-y^3+4x^2 y+y^2+fx+ky; V_triple=x^6-x^4+x^2+2x^2 y+4y^4-y^2+fx+ky; V_heteroclinic=x^4+y^4-y^3+2x^2…
    Coefficients chosen so the decision graph matches the desired class; small changes alter trajectory shapes and signal thresholds.
  • Inverse temperature beta = 2N, where N is the number of genes simulated
    Set to ensure e^beta is much larger than the number of patterns for convergence; not fitted to data.
  • Lateral inhibition parameters A and a for spatial patterning = a shared across landscapes; A differs by landscape (Fig. S6)
    Chosen so that landscapes with different basin depths respond to Notch signaling; not fitted to data.
assumptions (6)
  • domain assumption Cell fates can be modeled as stable fixed points (attractors) in gene expression space; progenitor states are transient attractors.
    Stated at the start of Section II A; the entire landscape interpretation rests on this.
  • domain assumption Developmental systems are gradient-like Morse-Smale systems, so fold and heteroclinic flips suffice to connect decision classes.
    Invoked from Rand et al. [23] and Raju and Siggia [24] in Section II B; not proved in this paper.
  • standard math A two-dimensional normal form captures the qualitative behavior of a binary cell fate decision.
    SI Section S1 A 1 asserts that local bifurcation models are universal, so 2D landscapes suffice for a binary decision.
  • domain assumption Projection onto the reference cell fate subspace (scTOP coordinates) discards only fate-irrelevant gene expression.
    Eqs. 4-7 and Methods A; the paper states x_perp is ignored. If the basis is incomplete, this assumption fails.
  • ad hoc to paper The chosen typical topographies and signal schedules produce trajectories representative of their decision classes.
    Main text: topology does not fully determine trajectories; SI B1: schedules chosen to emphasize class features. Without this, the signature-to-class mapping is not justified.
  • ad hoc to paper Clonal families with exactly two mature fates in the lineage tracing data correspond to bipotent decisions on a three-attractor landscape.
    Section II C1 selection criterion; multi-fate progenitors are excluded, which may bias inferred landscapes.
invented entities (1)
  • Transient AT1/AT2 mixed-state progenitor in the developing lung
    purpose: Explains the central high-density cluster between AT1 and AT2 fates as an attractor stabilized until birth and destabilized by air-breathing.
    The only evidence is the cluster in the same Zepp et al. data used to motivate the hypothesis; the paper proposes in vitro culture experiments that could provide independent evidence (Discussion).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Collective gene dynamics leave signatures of decision landscapes in cell fate coordinates." pith.science (2026). https://pith.science/paper/32S76LNT

@misc{pith2026250604219,
  author       = {Pith},
  title        = {Pith review of: Collective gene dynamics leave signatures of decision landscapes in cell fate coordinates},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/32S76LNT}},
  note         = {Machine review of arXiv:2506.04219}
}
read the original abstract

Multicellular organisms contain a wide variety of highly specialized cell types. The consistency and robustness of developmental trajectories suggest that complex gene regulatory networks effectively act as low-dimensional cell fate landscapes. Prior work inspired by dynamical systems theory argues that cell fate transitions fall into universal decision-making classes, but the theory connecting these geometric landscapes to high-dimensional gene expression space is still in its infancy. Here, we introduce a phenomenological model that identifies experimental signatures of decision-making classes in single-cell RNA-sequencing time-series data. The model combines low-dimensional gradient-like dynamics with high-dimensional Hopfield networks to capture the interplay between cell fate, gene expression, and signaling. We apply the framework to experimental mouse data on maturing lung alveolar cells and lineage-traced hematopoietic differentiation and show that the measured cell fate dynamics are consistent with developmental landscapes containing intermediate progenitors and saddle points. We further show that the framework can be used to understand spatial patterning and cell fate organization, focusing on Notch signaling in lung airways. Together, these results provide evidence that collective transcriptomic dynamics carry signatures of landscape features associated with universal decision-making classes.

Figures

Figures reproduced from arXiv: 2506.04219 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Inheritance Entropy: A Model-Independent Method to Probe the Hereditary Structure of Cell Lineage Trees

    physics.bio-ph 2025-10 conditional novelty 6.0 of 10

    Cell-cycle exit in human bone marrow stromal cells is statistically clustered along lineage branches, indicating heritable (epigenetic) regulation, as detected by a new lineage-topology entropy.

Reference graph

Works this paper leans on

96 extracted references · 75 canonical work pages · cited by 1 Pith paper

  1. [1]

    These blood cells originate from hematopoietic stem and progenitor cells (HSPCs), which have the capacity to differentiate into many blood cell fates

    Cell fate decisions in hematopoiesis Blood requires frequent replenishment, so a wide range of blood cells are regularly created in bone marrow. These blood cells originate from hematopoietic stem and progenitor cells (HSPCs), which have the capacity to differentiate into many blood cell fates. They are responsible for maintaining stable populations of bl...

  2. [2]

    Different transitions between these types have been established in homeostasis, injury, and transplant [54]

    Cell fate decisions in the developing lung With a variety of specialized cell types, the lung provides a rich system for studying cell fate decisions. Different transitions between these types have been established in homeostasis, injury, and transplant [54]. In this section, we explore the case of alveolar maturation in murine develop- ment. In the mouse...

  3. [3]

    Wolpert, C

    L. Wolpert, C. Tickle, and A. M. Arias,Principles of development(Oxford University Press, USA, 2015)

    2015

  4. [4]

    Debat and P

    V. Debat and P. David, Trends in ecology & evolution16, 555 (2001)

    2001

  5. [5]

    Wagner,Robustness and evolvability in living systems(Princeton university press, 2013)

    A. Wagner,Robustness and evolvability in living systems(Princeton university press, 2013)

    2013

  6. [6]

    Takahashi and S

    K. Takahashi and S. Yamanaka, cell126, 663 (2006)

    2006

  7. [7]

    Takahashi and S

    K. Takahashi and S. Yamanaka, Nature reviews Molecular cell biology17, 183 (2016)

    2016

  8. [8]

    Srivastava and N

    D. Srivastava and N. DeWitt, Cell166, 1386 (2016)

    2016

Show all 96 references
  1. [9]

    R. A. Shivdasani, H. Clevers, and F. J. de Sauvage, Science371, 784 (2021)

    2021

  2. [10]

    Trapnell, Genome research25, 1491 (2015)

    C. Trapnell, Genome research25, 1491 (2015)

    2015

  3. [11]

    A. A. Kolodziejczyk, J. K. Kim, V. Svensson, J. C. Marioni, and S. A. Teichmann, Molecular cell58, 610 (2015)

    2015

  4. [12]

    Svensson, R

    V. Svensson, R. Vento-Tormo, and S. A. Teichmann, Nature protocols13, 599 (2018)

    2018

  5. [13]

    Jones, J

    R. Jones, J. Karkanias, M. Krasnow, A. Pisco, S. Quake, J. Salzman, N. Yosef, B. Bulthaup, P. Brown, W. Harper,et al., Science376, 6594 (2022)

    2022

  6. [14]

    X. Han, R. Wang, Y. Zhou, L. Fei, H. Sun, S. Lai, A. Saadatpour, Z. Zhou, H. Chen, F. Ye,et al., Cell 172, 1091 (2018)

    2018

  7. [15]

    C. Qiu, B. K. Martin, I. C. Welsh, R. M. Daza, T.-M. Le, X. Huang, E. K. Nichols, M. L. Taylor, O. Fulton, D. R. O’Day,et al., Nature626, 1084 (2024). 24

    2024

  8. [16]

    L¨ ahnemann, J

    D. L¨ ahnemann, J. K¨ oster, E. Szczurek, D. J. McCarthy, S. C. Hicks, M. D. Robinson, C. A. Vallejos, K. R. Campbell, N. Beerenwinkel, A. Mahfouz,et al., Genome biology21, 1 (2020)

    2020

  9. [17]

    C. H. Waddington,The strategy of the genes(Routledge, 2014)

    2014

  10. [18]

    Huang, Bioessays34, 149 (2012)

    S. Huang, Bioessays34, 149 (2012)

    2012

  11. [19]

    Thom,Structural stability and morphogenesis(CRC press, 2018)

    R. Thom,Structural stability and morphogenesis(CRC press, 2018)

    2018

  12. [20]

    J. Wang, K. Zhang, L. Xu, and E. Wang, Proceedings of the National Academy of Sciences108, 8257 (2011)

    2011

  13. [21]

    J. E. Ferrell, Current biology22, R458 (2012)

    2012

  14. [22]

    Moris, C

    N. Moris, C. Pina, and A. M. Arias, Nature Reviews Genetics17, 693 (2016)

    2016

  15. [23]

    S´ aez, R

    M. S´ aez, R. Blassberg, E. Camacho-Aguilar, E. D. Siggia, D. A. Rand, and J. Briscoe, Cell Systems 13, 12 (2022)

    2022

  16. [24]

    Wang, Journal of Biological Physics48, 1 (2022)

    J. Wang, Journal of Biological Physics48, 1 (2022)

    2022

  17. [25]

    D. A. Rand, A. Raju, M. S´ aez, F. Corson, and E. D. Siggia, Proceedings of the National Academy of Sciences118, e2109729118 (2021)

    2021

  18. [26]

    Raju and E

    A. Raju and E. D. Siggia, Development, Growth & Differentiation65, 245 (2023), https://onlinelibrary.wiley.com/doi/pdf/10.1111/dgd.12855

    2023 doi

  19. [27]

    Camacho-Aguilar, A

    E. Camacho-Aguilar, A. Warmflash, and D. A. Rand, PLoS computational biology17, e1009034 (2021)

    2021

  20. [28]

    D. J. Cislo, M. J. Del´ as, J. Briscoe, and E. D. Siggia, Proceedings of the National Academy of Sciences 122, e2521762122 (2025)

    2025

  21. [29]

    Mochulska and P

    V. Mochulska and P. Fran¸ cois, Proceedings of the National Academy of Sciences122, e2514508122 (2025)

    2025

  22. [30]

    S´ aez, J

    M. S´ aez, J. Briscoe, and D. A. Rand, Interface focus12, 20220002 (2022)

    2022

  23. [31]

    Yampolskaya, M

    M. Yampolskaya, M. J. Herriges, L. Ikonomou, D. N. Kotton, and P. Mehta, Development150, dev201873 (2023)

    2023

  24. [32]

    Grishechkin, A

    A. Grishechkin, A. Mukherjee, and O. Karin, PRX Life3, 043007 (2025)

    2025

  25. [33]

    M. Hu, D. Krause, M. Greaves, S. Sharkis, M. Dexter, C. Heyworth, and T. Enver, Genes & develop- ment11, 774 (1997)

    1997

  26. [34]

    Olsson, M

    A. Olsson, M. Venkatasubramanian, V. K. Chaudhri, B. J. Aronow, N. Salomonis, H. Singh, and H. L. Grimes, Nature537, 698 (2016)

    2016

  27. [35]

    J. J. Hopfield, Proceedings of the national academy of sciences79, 2554 (1982)

    1982

  28. [36]

    Krotov and J

    D. Krotov and J. J. Hopfield, Advances in neural information processing systems29(2016)

    2016

  29. [37]

    Yampolskaya and P

    M. Yampolskaya and P. Mehta, Annual Review of Biophysics55(2025)

    2025

  30. [38]

    A. H. Lang, H. Li, J. J. Collins, and P. Mehta, PLoS computational biology10, e1003734 (2014)

    2014

  31. [39]

    A. T. Fard, S. Srihari, J. C. Mar, and M. A. Ragan, NPJ systems biology and applications2, 1 (2016)

    2016

  32. [40]

    S. T. Pusuluri, A. H. Lang, P. Mehta, and H. E. Castillo, Physical Biology15, 016001 (2017)

    2017

  33. [41]

    Guo and J

    J. Guo and J. Zheng, Bioinformatics33, i102 (2017)

    2017

  34. [42]

    Ikonomou, M

    L. Ikonomou, M. J. Herriges, S. L. Lewandowski, R. Marsland III, C. Villacorta-Martin, I. S. Caballero, D. B. Frank, R. M. Sanghrajka, K. Dame, M. M. Ka´ ndu la,et al., Nature communications11, 635 (2020)

    2020

  35. [43]

    Smart and A

    M. Smart and A. Zilman, Cell Reports Physical Science4(2023)

    2023

  36. [44]

    Karin, Development151, dev202997 (2024)

    O. Karin, Development151, dev202997 (2024)

    2024

  37. [45]

    N. E. Boukacem, A. Leary, R. Th´ eriault, F. Gottlieb, M. Mani, and P. Fran¸ cois, Physical Review Research6, 033098 (2024)

    2024

  38. [46]

    Souza and P

    H. Souza and P. Mehta, bioRxiv , 2026 (2026)

    2026

  39. [47]

    S. H. Strogatz,Nonlinear dynamics and chaos: with applications to physics, biology, chemistry, and engineering (studies in nonlinearity), Vol. 1 (Westview press, 2001)

    2001

  40. [48]

    J. J. Hopfield, Proceedings of the national academy of sciences81, 3088 (1984)

    1984

  41. [50]

    Yampolskaya and P

    M. Yampolskaya and P. Mehta, inAssociative Memory{\&}Hopfield Networks in 2023(2023)

    2023

  42. [51]

    Kanter and H

    I. Kanter and H. Sompolinsky, Physical Review A35, 380 (1987)

    1987

  43. [52]

    C. Zeeman, inStructural Stability, the Theory of Catastrophes, and Applications in the Sciences: Pro- ceedings of the Conference Held at Battelle Seattle Research Center 1975(Springer, 2006) pp. 263–327

    1975

  44. [53]

    Bialek,Biophysics: searching for principles(Princeton University Press, 2012)

    W. Bialek,Biophysics: searching for principles(Princeton University Press, 2012). 25

    2012

  45. [54]

    Newhouse and M

    S. Newhouse and M. Peixoto, Asterisque31, 15 (1976)

    1976

  46. [55]

    Weinreb, A

    C. Weinreb, A. Rodriguez-Fraticelli, F. D. Camargo, and A. M. Klein, Science367, eaaw3381 (2020)

    2020

  47. [56]

    Alysandratos, M

    K.-D. Alysandratos, M. J. Herriges, and D. N. Kotton, Annual review of physiology83, 529 (2021)

    2021

  48. [57]

    Hawkins, S

    F. Hawkins, S. A. Rankin, D. N. Kotton, and A. M. Zorn, The genetic programs regulating embryonic lung development and induced pluripotent stem cell differentiation, inFetal and Neonatal Lung De- velopment: Clinical Correlates and Technologies for the Future(Cambridge Universi...

    2016

  49. [58]

    J. A. Zepp, M. P. Morley, C. Loebel, M. M. Kremp, F. N. Chaudhry, M. C. Basil, J. P. Leach, D. C. Liberti, T. K. Niethamer, Y. Ying,et al., Science371, eabc3172 (2021)

    2021

  50. [59]

    M. J. Herriges, M. Yampolskaya, B. R. Thapa, J. Lindstrom-Vautrin, F. Wang, C.-L. Na, L. Ma, M. M. Montminy, J. Huang, C. Villacorta-Martin, P. Mehta, and D. N. Kotton,Durable alveolar engraftment of PSC-derived lung epithelial cells into immunocompetent mice, preprint (Cell B...

    2022

  51. [60]

    J. R. Rock, X. Gao, Y. Xue, S. H. Randell, Y.-Y. Kong, and B. L. Hogan, Cell stem cell8, 639 (2011)

    2011

  52. [61]

    J. A. Whitsett and V. V. Kalinichenko, Cell Stem Cell8, 597 (2011)

    2011

  53. [62]

    D. B. Br¨ uckner and G. Tkaˇ cik, Proceedings of the National Academy of Sciences121, e2322326121 (2024)

    2024

  54. [63]

    J. R. Rock, M. W. Onaitis, E. L. Rawlins, Y. Lu, C. P. Clark, Y. Xue, S. H. Randell, and B. L. Hogan, Proceedings of the National Academy of Sciences106, 12771 (2009)

    2009

  55. [64]

    Pardo-Saganta, B

    A. Pardo-Saganta, B. M. Law, P. R. Tata, J. Villoria, B. Saez, H. Mou, R. Zhao, and J. Rajagopal, Cell stem cell16, 184 (2015)

    2015

  56. [65]

    Goetz, H

    A. Goetz, H. Akl, and P. Dixit, Elife12, RP87747 (2024)

    2024

  57. [66]

    R. Wang, P. Zhang, J. Wang, L. Ma, W. E, S. Suo, M. Jiang, J. Li, H. Chen, H. Sun,et al., Nucleic acids research51, 501 (2023)

    2023

  58. [67]

    M. J. Herriges, M. Yampolskaya, B. R. Thapa, J. Lindstrom-Vautrin, F. Wang, J. Huang, C.-L. Na, L. Ma, M. M. Montminy, P. Bawa,et al., Cell Stem Cell30, 1217 (2023)

    2023

  59. [68]

    N. M. Negretti, E. J. Plosa, J. T. Benjamin, B. A. Schuler, A. C. Habermann, C. S. Jetter, P. Gulleman, C. Bunn, A. N. Hackett, M. Ransom,et al., Development148, dev199512 (2021). 1 S1. SUPPLEMENT AL INFORMA TION CONTENTS S1. Supplemental Information 1 A. Mathematical details 1

    2021

  60. [69]

    Choice of potential 1

  61. [70]

    Landscape coordinate transformation 2

  62. [71]

    Correlated cell types and invariant transformations 5

  63. [72]

    Potentials over time 7

  64. [73]

    Simulation details 8

    Spatial patterning 7 B. Simulation details 8

  65. [74]

    Signaling dynamics 9

  66. [75]

    Inverse temperature 11

  67. [76]

    Limitations of the model 12 References 12 A

    Robustness to update schedule 12 C. Limitations of the model 12 References 12 A. Mathematical details

  68. [77]

    As explained in Randet al.[1], two of the simplest bifurcations with three attractors are the binary cusp and the heteroclinic flip

    Choice of potential In choosing possible landscapes, we only considered landscapes containing three attractors. As explained in Randet al.[1], two of the simplest bifurcations with three attractors are the binary cusp and the heteroclinic flip. However, in the case of AT1/AT2 ...

  69. [78]

    In our model, alignment with each cell type is given its own axis, set by the Hopfield-inspired order parametersm µ

    Landscape coordinate transformation The normal forms corresponding to each decision-making classes take the form of a 2-dimensional potential,V(˜x,˜y). In our model, alignment with each cell type is given its own axis, set by the Hopfield-inspired order parametersm µ. To align...

  70. [79]

    Double cusp B

    Scaling with matrixSsuch that the triangle formed bya 0, a1, a2 becomes equilateral with side lengths √ 2.S=   √ 2 2x1 0 0 0 0 √ 3/2 y0−y1 0 0 0 0 1 0 0 0 0 1   4 A. Double cusp B. Triple cusp C. Heteroclinic /f_lip Simulation time Start End FIG. S2. Simulations of c...

  71. [80]

    Translating thea 0, a1, a2 triangle with matrixT 1 such that the midpoint of the bottom side is a distance √ 2 2 away from the origin.T1 =   1 0 0 0 0 1 0− √ 2 2 − y1 √ 3/2 y0−y1 0 0 1 0 0 0 0 1   5

  72. [81]

    Rotating about the third axis by an angle of 3π 4 with matrix R1 =   cos 3π 4 −sin 3π 4 0 0 sin 3π 4 cos 3π 4 0 0 0 0 1 0 0 0 0 1  

  73. [82]

    For trans- forming from cell-type spacem µ to landscape space ˜x,˜y, we defineH=K −1.h x,y µ denote the entries ofHsuch that ˜x= P µ hx µmµ and ˜y= P µ hy µmµ

    Rotating by an angle of arcsin q 2 3 about the line that crosses through (1, 0, 0, 1) and (0, 1, 0, 1) with matrixM=T −1X −1Y −1ZY XT.T, X, Y, Zare defined as follows: T=   1 0 0−1 0 1 0 0 0 0 1 0 0 0 0 1   X=   1 0 0 0 0 0−1 0 0 1 0 0 0 0 0 1   Y=   √ 2 2 0 √ ...

  74. [83]

    Correlated cell types and invariant transformations Cell types are often highly correlated, especially if they have common progenitors. Kanter and Sompolinsky [3] proposed an attractor network storage method which allowed for many highly- correlated attractors to coexist, whic...

  75. [84]

    In other words, the update rule of the Hopfield model causes this energy function to decrease

    Potentials over time It’s well-known that the Hopfield model has an associated energy landscape which acts as the Lyapunov function of the system [5]. In other words, the update rule of the Hopfield model causes this energy function to decrease. For the modern Hopfield model, ...

  76. [85]

    Each cell experiences a landscape that is biased towards one fate or another by its neighboring cells

    Spatial patterning In Section II D, to simulate cells with a spatial pattern of alternating cell fates, we model the system as a line of cells. Each cell experiences a landscape that is biased towards one fate or another by its neighboring cells. To identify the effects of dif...

  77. [86]

    The values for the signaling parameters (k, f) were varied as follows

    Signaling dynamics For Figure 2, the potentials were plotted in Mathematica using the equations shown. The values for the signaling parameters (k, f) were varied as follows. For the landscape with a triple cusp, the 10 system starts withk= 0, f= 0. The signaling parameterkis i...

  78. [87]

    For the modern Hopfield network to converge to the correct pattern, it is necessary that the temperature is low enough such thate β ≫P, wherePis the number of stored patterns

    Inverse temperature For all simulations except when explicitly stated otherwise, the inverse temperatureβin the softmax of the update rule was set toβ= 2N, whereNis the number of genes simulated. For the modern Hopfield network to converge to the correct pattern, it is necessa...

  79. [88]

    The simulations were initiated at the initial attractor state with noise added to the gene expression:x i(t= 0) =ξ 0i +z i

    Noise Noise was added to the simulation in Figures S2 and 3 to imitate the noise in scRNA-seq and facilitate comparison between simulated trajectories and data. The simulations were initiated at the initial attractor state with noise added to the gene expression:x i(t= 0) =ξ 0...

  80. [89]

    However, this does not reflect biological reality

    Robustness to update schedule In our model, all genes are updated simultaneously. However, this does not reflect biological reality. To reflect the asynchronous nature of gene regulation, we can randomly update some genes at each simulation step and not others. The model is ro...

  81. [90]

    D. A. Rand, A. Raju, M. S´ aez, F. Corson, and E. D. Siggia, Geometry of gene regulatory dynamics, Proceedings of the National Academy of Sciences118, e2109729118 (2021)

    2021

  82. [91]

    S´ aez, R

    M. S´ aez, R. Blassberg, E. Camacho-Aguilar, E. D. Siggia, D. A. Rand, and J. Briscoe, Statistically de- rived geometrical landscapes capture principles of decision-making dynamics during cell fate transitions, Cell Systems13, 12 (2022)

    2022

  83. [92]

    Kanter and H

    I. Kanter and H. Sompolinsky, Associative recall of memory without errors, Physical Review A35, 380 (1987). 13 All genes update every step 1 in 5 genes update every step 1 in 9 genes update every step FIG. S7. The dynamical system is robust to changes in update schedule. All t...

    1987

  84. [93]

    Chaudhry, J

    H. Chaudhry, J. Zavatone-Veth, D. Krotov, and C. Pehlevan, Long sequence hopfield memory, Advances in Neural Information Processing Systems36(2024)

    2024

  85. [94]

    Krotov and J

    D. Krotov and J. Hopfield, Large associative memory problem in neurobiology and machine learning, arXiv preprint arXiv:2008.06996 (2020). 14 A. B. C. Days in vitro Day 2 Day 6 Developmental time E12.5 P42 Days in vitro Day 2 Day 6 FIG. S8. The same data as in Figure 4, shown a...

    2020 arXiv

  86. [95]

    M. J. Herriges, M. Yampolskaya, B. R. Thapa, J. Lindstrom-Vautrin, F. Wang, J. Huang, C.-L. Na, L. Ma, M. M. Montminy, P. Bawa,et al., Durable alveolar engraftment of psc-derived lung epithelial cells into immunocompetent mice, Cell Stem Cell30, 1217 (2023)

    2023

  87. [96]

    N. M. Negretti, E. J. Plosa, J. T. Benjamin, B. A. Schuler, A. C. Habermann, C. S. Jetter, P. Gulleman, C. Bunn, A. N. Hackett, M. Ransom,et al., A single-cell atlas of mouse lung development, Development 148, dev199512 (2021)

    2021

  88. [97]

    S. C. Hicks, F. W. Townes, M. Teng, and R. A. Irizarry, Missing data and technical variability in single-cell rna-sequencing experiments, Biostatistics19, 562 (2018)

    2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.