acceptodds
Under review as a conference paper at ICLR 2027

SpudLenia: Coupling Synthetic Protocell Hydrodynamics with Continuous Cellular Automata for Autonomous Morphogenesis

Abstract

Achieving autonomous cytokinesis in continuous fields without active cytoskeletal machinery remains an open challenge across artificial life, biophysics, and self-organizing systems. While biological eukaryotes rely on ATP-driven actin-myosin contractile rings, recent wet-lab synthetic protocells (SpudCells) demonstrate that membrane-associated protein crowding can induce furrow constriction; however, physical prototypes remain bottlenecked by external chemical triggers and mechanical extrusion. Here, we introduce SpudLenia, an in silico differentiable continuous cellular automata framework that tests whether motor-free cytokinesis can emerge autonomously in a 2D reaction-convolution field. SpudLenia operates on multi-channel fields representing lipid bilayers, crowding proteins, and nutrient liposomes, coupled via multi-scale spatial convolution kernels and channel-specific growth activations. We hypothesize that localized steric crowding imposes an effective spatial growth penalty, and test whether this single inductive bias is sufficient for cytokinesis. By parameterizing a localized growth-braking operator that quenches lipid expansion along steep crowding gradients, SpudLenia produces midline furrow constriction with a characteristic constriction timescale consistent with experimental SpudCell velocities (; Adamala et al., 2026) and culminating in complete topological membrane scission into stable secondary vesicles—demonstrating in silico the origin of the incomplete fission bottleneck observed in wet-lab prototypes. We characterize the open-system mass flux balance, identifying a stable homeostatic attractor that bounds growth against pathological bloating. Implemented as a differentiable 2D convolutional field solver running at 1.2 ms per step ( lattice on consumer GPUs, e.g., Apple M1 Pro), SpudLenia achieves a 94.6% fission success rate across 500 trials perturbed over initial protocell geometry, nutrient placement, and protein noise, compared to 0% without the growth-braking operator. SpudLenia provides a principled continuous substrate for modeling decentralized morphogenesis and predicting division regimes in synthetic protocells; code and an interactive web visualizer are publicly available.

Primary area:
applications to physical sciences (physics, chemistry, biology, etc.)
Keywords:
Neural Cellular Automata, Continuous Cellular Automata, Lenia, Differentiable Physics, Morphogenesis, Synthetic Biology, Cytokinesis, Quality-Diversity, MAP-Elites, Self-Organization, Dynamical Systems

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_VfceOpenReview ↗
Rating
1
Confidence
1
Informativeness
2
Presentation
1
Soundness
1
Read review
Paper summary

The paper introduces SpudLenia, a differentiable continuous cellular automata framework for simulating motor-free protocell division. It couples lipid, protein-crowding, and nutrient fields through a localized growth-braking mechanism to induce membrane constriction and scission. The authors report a 94.6% division success rate across 500 perturbed simulations and use ablation studies to assess the contributions of individual components.

Critical strengths

The paper couples lipid, protein-crowding, and nutrient fields within continuous cellular automata, providing an interpretable computational framework for exploring motor-free protocell division.

Critical weaknesses
  1. Equation (7) omits terms from the governing dynamics, and Proposition 1 does not establish global stability. These inconsistencies undermine the central claim of homeostatic mass regulation.
  2. The growth-braking operator lacks a convincing biophysical derivation. Its necessity within the constructed model does not establish it as a physical mechanism of protocell division.
  3. The experiments lack precise perturbation settings, scission criteria, and uncertainty estimates. The claimed agreement with experimental velocities also lacks a simulation-to-physical-unit mapping.
  4. The paper claims gradient-based optimization and MAP-Elites exploration without presenting quantitative results, leaving its optimization capabilities and morphological diversity insufficiently substantiated.
Questions for authors

Please refer to weaknesses.

Reviewer_AnqBOpenReview ↗
Rating
2
Confidence
1
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper introduces SpudLenia, a differentiable continuous cellular automata framework for simulating autonomous protocell division without cytoskeletal machinery. The model represents lipid membranes, protein crowding, and nutrient liposomes using three coupled continuous fields, whose dynamics are governed by spatial convolution kernels and reaction-diffusion equations. Its central mechanism is a localized growth-braking operator that suppresses lipid growth in regions exhibiting strong protein-crowding gradients, with the goal of inducing cleavage-furrow formation and membrane scission.

Critical strengths

The paper addresses an intriguing problem at the intersection of artificial life, differentiable simulation, and synthetic biology. Investigating minimal computational mechanisms that can produce division-like behavior in continuous fields is a potentially valuable research direction.

The framework uses a relatively compact parameterization with explicit fields for lipids, proteins, and nutrients. Its convolution-based implementation is computationally efficient and differentiable, making it potentially suitable for interactive exploration and parameter optimization.

The paper goes beyond presenting isolated simulations by including parameter sweeps, component ablations, and an analysis of different qualitative dynamical regimes. These are useful starting points for understanding how the proposed model behaves.

Critical weaknesses
  1. The main experimental visualization appears inconsistent with the claimed cytokinesis behavior. Figure 2(g) is described as demonstrating autonomous exponential narrowing of the cleavage furrow. However, the displayed curve for furrow width increases over time rather than decreasing. This directly contradicts the claimed constriction behavior, at least under the conventional interpretation of the plotted quantity as furrow diameter.

  2. The proposed model lacks sufficient biophysical grounding and validation. Although SpudLenia is presented as a biophysical model, its dynamics are primarily constructed from phenomenological growth functions and heuristic coupling terms. In particular, the localized growth-braking operator is introduced as a proxy for steric protein crowding. However, the derivation in Appendix A does not establish a complete mechanical correspondence between the Canham–Helfrich membrane energy, steric forces, and the proposed lipid-density reaction term. The model does not explicitly represent membrane curvature dynamics, membrane tension, pressure differences, hydrodynamics, or the three-dimensional geometry of lipid bilayers. These processes are important for real membrane deformation and scission.

  3. Novelty is overstated relative to existing continuous cellular automata research. The paper claims that classical Lenia solitons cannot divide and presents the proposed mechanism as resolving a fundamental limitation of continuous cellular automata. However, prior work on Lenia and related continuous cellular automata has already explored self-reproduction, pattern splitting, and populations of localized structures.

  4. Experimental validation is insufficient to support the reported success rates and mechanistic conclusions. The main quantitative evaluation consists of 500 simulations under perturbed initial conditions and several parameter ablations. However, the paper does not clearly specify the perturbation distributions, the operational definition of successful fission, the connectivity criterion used to detect topological separation, or the statistical uncertainty of the reported rates. More importantly, eliminating the proposed growth-braking term and observing the disappearance of fission does not establish that the term captures a physically meaningful biological mechanism. It only demonstrates that the selected simulator depends on this explicitly introduced term to produce the targeted behavior.

  5. The claimed differentiable optimization and quality-diversity contributions are not sufficiently substantiated.

Questions for authors

See weakness above‘

Reviewer_TtTgOpenReview ↗
Rating
1
Confidence
2
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

The paper built a computational model that simulate the cell dividing. The authors focus on the division caused by protein crowding rather than complex mechanistic model. They proposes an three-channel continuous cellular automaton for protocell division using a protein-gradient-dependent growth penalty.

Critical strengths
  1. The biological question that the authors aim to solve: modeling cell division from simple mechanism of protein crowding is important.
  2. The authors proposed an three-channel continuous cellular automaton for protocell division, which is innovative in simulation modeling.
  3. The model is interpretable, with parameter clearly defined based on the biological role.
Critical weaknesses
  1. There are some issues with the derived equations. The growth and neck-braking contributions that are omitted from Eq. (7) without explaining. Proposition 1 therefore does not follow from the stated dynamics.
  2. Figure 2g shows the width is increasing, but it should be narrowing instead. The authors should check on the discrepancy.
  3. More details should be provided with the gradient-based parameter optimization. There is no clearly defined differentiable objective for gradient descent. In addition, training protocol, or quantitative optimization results are also missing. The MAP-Elites experiments do not establish the utility of gradients. Please provide a concrete optimization experiment or narrow the claim to the simulator’s differentiability.
  4. The novelty relative to Expanded Lenia needs clarification. Although the discussion refers specifically to classical Lenia, Chan (2020) already reports multichannel extensions and self-replication through repeated division. The authors should clarify what the proposed crowding-dependent growth penalty contributes beyond Expanded Lenia and provide supporing experiments.
Questions for authors

See the weakness.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.