pith. sign in

arxiv: 2607.01749 · v1 · pith:UOTJEKNJnew · submitted 2026-07-02 · 🧬 q-bio.QM · physics.bio-ph· stat.ML

Identifiability Limits of Physics-Informed Inference for Spatial Stochastic Dynamics from Static Snapshots

Rujie Gu , Ray Zirui Zhang , Christopher E. Miles This is my paper

Pith reviewed 2026-07-03 02:27 UTC · model grok-4.3

classification 🧬 q-bio.QM physics.bio-phstat.ML
keywords structural identifiabilityphysics-informed machine learningspatial stochastic dynamicsstatic snapshotspoint sourcedistributed sourcegene expressioninference limits
0
0 comments X

The pith

Distributed sources are non-identifiable from static snapshots, but a point source restores identifiability.

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

The paper asks when a static spatial pattern of molecules can identify spatially varying diffusivity, creation, destruction, and boundary exchange. A structural identifiability analysis shows that distributed sources are non-identifiable, whereas a point source such as a transcription site can restore identifiability. These limits are shaped by boundary conditions, spatial regularity of the dynamics, and the stochastic calculus convention. Physics-informed schemes can recover spatial heterogeneities from a single snapshot when the conditions are met.

Core claim

A structural identifiability analysis shows that distributed sources are non-identifiable, whereas a point source such as a transcription site can restore identifiability. These limits are further shaped by boundary conditions, the spatial regularity of the underlying dynamics, and even the stochastic calculus convention. Adapted physics-informed schemes demonstrate effective inference from a single snapshot.

What carries the argument

Structural identifiability analysis for physics-informed inference of spatial stochastic dynamics from static snapshots.

If this is right

  • Physics-informed inference can recover spatial heterogeneities from static data if a point source is present.
  • Modeling choices like boundary conditions and stochastic calculus affect what can be identified.
  • Careful identifiability analysis is required for meaningful interpretation of results from such inference.
  • Effective inference is possible from a single snapshot under identifiable conditions.

Where Pith is reading between the lines

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

  • The findings suggest that in gene expression studies, the presence of transcription sites is crucial for inferring dynamics from snapshots.
  • Similar identifiability limits may apply to other spatial biological processes without localized sources.
  • Adding more data types could potentially overcome the non-identifiability of distributed sources.

Load-bearing premise

The identifiability results depend on specific modeling choices including boundary conditions, spatial regularity of the dynamics, and the stochastic calculus convention used.

What would settle it

Numerical experiments showing unique recovery of distributed source parameters or a change in identifiability when switching the stochastic calculus convention would test the central claims.

Figures

Figures reproduced from arXiv: 2607.01749 by Christopher E. Miles, Ray Zirui Zhang, Rujie Gu.

Figure 1
Figure 1. Figure 1: Schematic of the inverse problem. Particles are produced by a source B(x), move with spatially varying diffusivity D(x), and degrade. At steady state they are observed only through a static snapshot of their positions, which form a spatial Poisson process with intensity u(x), random in both number and location. The inverse problem is to recover features of the underlying dynamics, such as D(x) and the sour… view at source ↗
Figure 2
Figure 2. Figure 2: Same-density diffuse-source alternative under Itô dynamics with Dirichlet boundaries. Here γ = 10, D1(x) = 1.5 + sin(4x), and b1(x) = 1 σ √ 2π exp − (x − 0.5)2 /2σ 2  with σ = 0.1; the source perturbation is b2 = b1 + 1. The compensating D2 = D1 + δD follows the construction in Theorem 3.1. Panels show (a) the common density, (b) the indistinguishable diffusivities D1 and D2, and (c) the alternative sourc… view at source ↗
Figure 3
Figure 3. Figure 3: Same-density Itô alternative under absorbing Dirichlet boundaries. Here D1(x) = 2 + 0.8 sin(2πx), z = 0.5, γ = 5, b (1) 0 = 10, and b (2) 0 = 14; the alternative D2 = D1 + δD follows the construction in Theorem 4.6. Panels show (a) the common density, (b) the indistinguishable diffusivities D1 and D2, with δD continuous but kinked at z, and (c) the alternative source strengths. 0 0.2 0.4 0.6 0.8 1 x 1.4 1.… view at source ↗
Figure 4
Figure 4. Figure 4: Same-density Itô alternative under reflecting (Neumann) boundaries. Here D1(x) = 2 + 0.8 sin(2πx), b0 = 10, z = 0.5, and γ = 5; the alternative D2 = D1 + δD uses the homogeneous mode C = 1.5 from Theorem 4.6. Panels show (a) the common density, (b) the indistinguishable diffusivities D1 and D2, with δD continuous but kinked at z, and (c) the unchanged source strength. Theorem 4.7 (Fickian point-source iden… view at source ↗
Figure 5
Figure 5. Figure 5: Same-density Fickian alternative under absorbing Dirichlet boundaries. Here D1(x) = 2 + 0.5 cos(4πx), z = 0.5, γ = 5, b (1) 0 = 10, and b (2) 0 = 14; the alternative D2 = D1 + δD follows the construction in Theorem 4.7. Panels show (a) the common density, with the source kink at z, (b) the indistinguishable diffusivities D1 and D2, with δD continuous but kinked at z, and (c) the alternative source strength… view at source ↗
Figure 6
Figure 6. Figure 6: Recovery under Itô dynamics with Dirichlet boundaries, for Dtrue(x) = 0.10 + 0.05 sin(8πx) at source strengths b0 = 250 (a, b) and b0 = 1000 (c, d). In panels (a, c) the dotted black curve is the true D(x), the colored curves are the per-method median recovered D(x), and the shaded bands span the 10th to 90th percentile over 100 trials. In panels (b, d) the bars are the distribution of the inferred source … view at source ↗
Figure 7
Figure 7. Figure 7: Recovery under Fickian dynamics with Neumann boundaries, for the high-frequency target Dtrue(x) = 0.10 + 0.05 sin(8πx) at b0 = 1000. Even at this source strength only DTO tracks the oscillations of D(x), while PINN and BiLO smooth them out. Plotting conventions follow [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Recovery under Fickian dynamics with Neumann boundaries for the lower-frequency target Dtrue(x) = 0.10 + 0.05 sin(6πx) at b0 = 250 (a, b) and b0 = 1000 (c, d). All three methods recover the oscillations, and the reconstruction tightens as b0 increases. Plotting conventions follow [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Same-density alternative for Itô diffusion with unknown Robin permeabilities and fixed source strength. Here D1(x) = 2 + 0.5 sin(2πx), b0 = 10, z = 0.5, γ = 5, and κ0 = κL = 2; the perturbation uses flux offset c = 0.6 and follows the construction in Theorem 6.1. Panels show (a) the common density, (b) the indistinguishable diffusivities D1 and D2, (c) the unchanged point source, and (d) the alternative en… view at source ↗
read the original abstract

Despite increasing scale and resolution, many biological measurements remain destructive, revealing only spatial information rather than the dynamics it encodes. By combining flexible representations with mechanistic constraints, physics-informed machine learning offers a promising route to inferring these dynamics from static snapshots. Motivated by subcellular imaging of gene expression, we ask when a static spatial pattern of molecules can identify spatially varying diffusivity, creation, destruction, and boundary exchange, and how different inference schemes perform on the task. A structural identifiability analysis shows that distributed sources are non-identifiable, whereas a point source such as a transcription site can restore identifiability. These limits are further shaped by seemingly innocuous modeling choices: the boundary conditions, the spatial regularity of the underlying dynamics, and even the stochastic calculus convention. We then adapt several physics-informed schemes, differing in how they represent the solution and enforce the governing equations, and demonstrate effective inference from a single snapshot. Physics-informed approaches can thus recover spatial heterogeneities of biological dynamics from static data, but their use should be accompanied and guided by careful identifiability analysis for meaningful interpretation of the results.

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

0 major / 3 minor

Summary. The manuscript conducts a structural identifiability analysis of physics-informed inference for spatially varying parameters (diffusivity, creation, destruction, boundary exchange) in stochastic dynamics models, using only static spatial snapshots as data. It establishes that distributed sources are non-identifiable while point sources (e.g., transcription sites) can restore identifiability; these limits depend explicitly on boundary conditions, spatial regularity of the dynamics, and the stochastic calculus convention. The authors then adapt multiple physics-informed schemes that differ in solution representation and equation enforcement, and demonstrate that they recover spatial heterogeneities from a single snapshot in the motivating application of subcellular gene-expression imaging.

Significance. If the identifiability results and inference demonstrations hold, the work supplies a needed cautionary framework for physics-informed machine learning in quantitative biology, where destructive measurements often yield only static spatial data. Explicitly mapping how seemingly innocuous modeling choices (boundary conditions, regularity, Itô vs. Stratonovich) alter identifiability is a concrete strength that can guide model selection. The successful single-snapshot inference examples under the identified conditions further increase practical utility for subcellular imaging studies.

minor comments (3)
  1. The abstract states that identifiability limits depend on the stochastic calculus convention, but the main text should include an explicit side-by-side comparison (e.g., in a dedicated subsection or table) of the two conventions applied to the same model to make the dependence transparent to readers.
  2. When the different physics-informed schemes are introduced, a brief tabular summary of how each represents the solution and enforces the governing equations would improve readability and allow direct comparison of their performance on the identifiability-restoring point-source case.
  3. Figure captions should explicitly note which boundary condition and stochastic calculus convention were used for each panel so that readers can immediately connect the visuals to the identifiability claims.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary and significance assessment of the manuscript, as well as the recommendation for minor revision. We appreciate the recognition that the identifiability results and single-snapshot inference demonstrations provide a useful cautionary framework for physics-informed methods in quantitative biology.

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper performs a structural identifiability analysis on a physics-informed inference setup for spatial stochastic dynamics, explicitly stating dependence on boundary conditions, spatial regularity, and stochastic calculus convention. No load-bearing steps reduce by construction to fitted parameters, self-citations, or renamed inputs; the central claims about non-identifiability of distributed sources and restoration by point sources follow from the stated modeling choices without self-referential closure. The derivation remains self-contained against external mathematical benchmarks for identifiability.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract provides no details on free parameters, axioms, or invented entities used in the analysis.

pith-pipeline@v0.9.1-grok · 5736 in / 1113 out tokens · 36402 ms · 2026-07-03T02:27:23.776079+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

66 extracted references · 58 canonical work pages · 2 internal anchors

  1. [1]

    Imaging Individual mRNA Molecules Using Multiple Singly Labeled Probes

    A. Raj et al. “Imaging Individual mRNA Molecules Using Multiple Singly Labeled Probes”.Nature Methods5.10 (Oct. 2008), pp. 877–879.doi:10.1038/nmeth.1253

    work page doi:10.1038/nmeth.1253 2008

  2. [2]

    Spatially Resolved, Highly Multiplexed RNA Profiling in Single Cells

    K. H. Chen et al. “Spatially Resolved, Highly Multiplexed RNA Profiling in Single Cells”.Science 348.6233 (Apr. 2015), aaa6090.doi:10.1126/science.aaa6090

    work page doi:10.1126/science.aaa6090 2015

  3. [3]

    Transcriptome-Scale Super-Resolved Imaging in Tissues by RNA seqFISH+

    C.-H. L. Eng et al. “Transcriptome-Scale Super-Resolved Imaging in Tissues by RNA seqFISH+”. Nature568.7751 (Apr. 2019), pp. 235–239.doi:10.1038/s41586-019-1049-y

    work page doi:10.1038/s41586-019-1049-y 2019

  4. [4]

    Subcellular mRNA Localization Patterns across Tissues Resolved with Spatial Transcriptomics

    R. Novoselsky et al. “Subcellular mRNA Localization Patterns across Tissues Resolved with Spatial Transcriptomics”.Nature Communications(Apr. 2026), pp. 1–36.doi:10.1038/s41467-026-72156-7

    work page doi:10.1038/s41467-026-72156-7 2026

  5. [5]

    Dynamics of RNA Localization to Nuclear Speckles Are Connected to Splicing Efficiency

    J. Wu et al. “Dynamics of RNA Localization to Nuclear Speckles Are Connected to Splicing Efficiency”. Science Advances10.42 (Oct. 16, 2024), eadp7727.doi:10.1126/sciadv.adp7727

    work page doi:10.1126/sciadv.adp7727 2024

  6. [6]

    GlobalAnalysisofmRNALocalizationRevealsaProminentRoleinOrganizingCellular Architecture and Function

    E.Lécuyeretal.“GlobalAnalysisofmRNALocalizationRevealsaProminentRoleinOrganizingCellular Architecture and Function”.Cell131.1 (Oct. 2007), pp. 174–187.doi:10.1016/j.cell.2007.08.003

    work page doi:10.1016/j.cell.2007.08.003 2007

  7. [7]

    In the Right Place at the Right Time: Visualizing and Understanding mRNA Localization

    A. R. Buxbaum, G. Haimovich, and R. H. Singer. “In the Right Place at the Right Time: Visualizing and Understanding mRNA Localization”.Nature Reviews Molecular Cell Biology16.2 (Feb. 2015), pp. 95–109.doi:10.1038/nrm3918

    work page doi:10.1038/nrm3918 2015

  8. [8]

    Constitutive Splicing and Economies of Scale in Gene Expression

    F. Ding and M. B. Elowitz. “Constitutive Splicing and Economies of Scale in Gene Expression”.Nature Structural & Molecular Biology26.6 (June 2019), pp. 424–432.doi:10.1038/s41594-019-0226-x

    work page doi:10.1038/s41594-019-0226-x 2019

  9. [9]

    Mechanism of mRNA Transport in the Nucleus

    D. Y. Vargas et al. “Mechanism of mRNA Transport in the Nucleus”.Proceedings of the National Academy of Sciences102.47 (Nov. 2005), pp. 17008–17013.doi:10.1073/pnas.0505580102

    work page doi:10.1073/pnas.0505580102 2005

  10. [10]

    Physics-Informed Machine Learning

    G. E. Karniadakis et al. “Physics-Informed Machine Learning”.Nature Reviews Physics3.6 (May 24, 2021), pp. 422–440.doi:10.1038/s42254-021-00314-5

    work page doi:10.1038/s42254-021-00314-5 2021

  11. [11]

    Scientific Machine Learning Through Physics–Informed Neural Networks: Where We Are and What’s Next

    S. Cuomo et al. “Scientific Machine Learning Through Physics–Informed Neural Networks: Where We Are and What’s Next”.Journal of Scientific Computing92.3 (Sept. 2022), p. 88.doi:10.1007/s10915- 022-01939-z. 26

    work page doi:10.1007/s10915- 2022

  12. [12]

    SolvingInverseStochasticProblemsfromDiscreteParticleObservationsUsingtheFokker– Planck Equation and Physics-Informed Neural Networks

    X.Chenetal.“SolvingInverseStochasticProblemsfromDiscreteParticleObservationsUsingtheFokker– Planck Equation and Physics-Informed Neural Networks”.SIAM Journal on Scientific Computing43.3 (Jan. 2021), B811–B830.doi:10.1137/20M1360153

    work page doi:10.1137/20m1360153 2021

  13. [13]

    Encoding Physics to Learn Reaction–Diffusion Processes

    C. Rao et al. “Encoding Physics to Learn Reaction–Diffusion Processes”.Nature Machine Intelligence 5.7 (July 2023), pp. 765–779.doi:10.1038/s42256-023-00685-7

    work page doi:10.1038/s42256-023-00685-7 2023

  14. [14]

    Biologically-Informed Neural Networks Guide Mechanistic Modeling from Sparse Experimental Data

    J. H. Lagergren et al. “Biologically-Informed Neural Networks Guide Mechanistic Modeling from Sparse Experimental Data”.PLOS Computational Biology16.12 (Dec. 1, 2020), e1008462.doi: 10.1371/ journal.pcbi.1008462

    2020

  15. [15]

    Lavery et al.Physics-Informed Neural Networks for Biological2D +t Reaction-Diffusion Systems

    W. Lavery et al.Physics-Informed Neural Networks for Biological2D +t Reaction-Diffusion Systems

  16. [16]

    Physics-Informed Neural Networks for Biological $2\mathrm{D}{+}t$ Reaction-Diffusion Systems

    arXiv:2604.18548 [cs.LG]. Preprint

    work page internal anchor Pith review Pith/arXiv arXiv

  17. [17]

    Systems Biology: Identifiability Analysis and Parameter Identification via Systems- Biology-Informed Neural Networks

    M. Daneker et al. “Systems Biology: Identifiability Analysis and Parameter Identification via Systems- Biology-Informed Neural Networks”.Computational Modeling of Signaling Networks. Ed. by L. K. Nguyen. New York, NY: Springer US, 2023, pp. 87–105.doi:10.1007/978-1-0716-3008-2_4

    work page doi:10.1007/978-1-0716-3008-2_4 2023

  18. [18]

    Position: Biology Is the Challenge Physics-Informed ML Needs to Evolve

    J. Martinelli. “Position: Biology Is the Challenge Physics-Informed ML Needs to Evolve”.Advances in Neural Information Processing Systems. Ed. by D. Belgrave et al. Vol. 38. Curran Associates, Inc., 2025.url: https://proceedings.neurips.cc/paper_files/paper/2025/file/af6c0a1b4a4faf1c 9900d2771bafd672-Paper-Position_Paper_Track.pdf

    2025

  19. [19]

    Inferring Stochastic Rates from Heterogeneous Snapshots of Particle Positions

    C. E. Miles et al. “Inferring Stochastic Rates from Heterogeneous Snapshots of Particle Positions”. Bulletin of Mathematical Biology86.6 (June 2024), pp. 74–104.doi:10.1007/s11538-024-01301-4

    work page doi:10.1007/s11538-024-01301-4 2024

  20. [20]

    Incorporating Spatial Diffusion into Models of Bursty Stochastic Transcription

    C. E. Miles. “Incorporating Spatial Diffusion into Models of Bursty Stochastic Transcription”.Journal of the Royal Society, Interface22.225 (Apr. 2025), p. 20240739.doi:10.1098/rsif.2024.0739

    work page doi:10.1098/rsif.2024.0739 2025

  21. [21]

    Y. Lin, H. Lin, and K. D. Welsher.Super-Resolving Particle Diffusion Heterogeneity in Porous Hydrogels via High-Speed 3D Active-Feedback Single-Particle Tracking Microscopy. Mar. 2025.doi:10.1101/ 2025.03.13.643103. Preprint

    2025

  22. [22]

    Detection of Diffusion Heterogeneity in Single Particle Tracking Trajectories Using a Hidden Markov Model with Measurement Noise Propagation

    P. J. Slator, C. W. Cairo, and N. J. Burroughs. “Detection of Diffusion Heterogeneity in Single Particle Tracking Trajectories Using a Hidden Markov Model with Measurement Noise Propagation”.PLOS ONE10.10 (Oct. 16, 2015), e0140759.doi:10.1371/journal.pone.0140759

    work page doi:10.1371/journal.pone.0140759 2015

  23. [23]

    ejpoleco.2025.102749 Farzanegan, M.R., Gutmann, J.,

    V. Kumar et al. “DiffMAP-GP: Continuous 2D Diffusion Maps from Particle Trajectories without Data Binning Using Gaussian Processes”.Biophysical Reports5.1 (Mar. 2025), p. 100194.doi:10.1016/j. bpr.2024.100194

    work page doi:10.1016/j 2025

  24. [24]

    Inferring Pointwise Diffusion Properties of Single Trajectories with Deep Learning

    B. Requena et al. “Inferring Pointwise Diffusion Properties of Single Trajectories with Deep Learning”. Biophysical Journal122.22 (Nov. 2023), pp. 4360–4369.doi:10.1016/j.bpj.2023.10.015

    work page doi:10.1016/j.bpj.2023.10.015 2023

  25. [25]

    Temporal Tissue Dynamics from a Spatial Snapshot

    J. Somer, S. Mannor, and U. Alon. “Temporal Tissue Dynamics from a Spatial Snapshot”.Nature 650.8101 (Feb. 2026), pp. 490–499.doi:10.1038/s41586-025-09876-1

    work page doi:10.1038/s41586-025-09876-1 2026

  26. [26]

    Inferring Stochastic Dynamics with Growth from Cross-Sectional Data

    S. Y. Zhang et al. “Inferring Stochastic Dynamics with Growth from Cross-Sectional Data”.The Thirty-ninth Annual Conference on Neural Information Processing Systems. Oct. 2025.url:https: //openreview.net/forum?id=MtdC1XS6RN

    2025

  27. [27]

    RNA Velocity of Single Cells

    G. La Manno et al. “RNA Velocity of Single Cells”.Nature560.7719 (Aug. 2018), pp. 494–498.doi: 10.1038/s41586-018-0414-6

    work page doi:10.1038/s41586-018-0414-6 2018

  28. [28]

    Optimal-Transport Analysis of Single-Cell Gene Expression Identifies Develop- mental Trajectories in Reprogramming

    G. Schiebinger et al. “Optimal-Transport Analysis of Single-Cell Gene Expression Identifies Develop- mental Trajectories in Reprogramming”.Cell176.4 (Feb. 2019), 928–943.e22.doi:10.1016/j.cell. 2019.01.006

    work page doi:10.1016/j.cell 2019

  29. [29]

    Identifiability and Reconstruction of Biochemical Reaction Networks from Population Snapshot Data

    E. Cinquemani. “Identifiability and Reconstruction of Biochemical Reaction Networks from Population Snapshot Data”.Processes6.9 (Aug. 2018), pp. 136–159.doi:10.3390/pr6090136

    work page doi:10.3390/pr6090136 2018

  30. [30]

    Guan et al.Identifying Drift, Diffusion, and Causal Structure from Temporal Snapshots

    V. Guan et al.Identifying Drift, Diffusion, and Causal Structure from Temporal Snapshots. 2024. arXiv: 2410.22729 [stat]. Preprint

    work page arXiv 2024

  31. [31]

    M. J. Simpson and M. J. Plank.When Do Trajectories Matter? Identifiability Analysis for Stochastic Transport Phenomena. 2026. arXiv:2604.15598 [nlin.CG]. Preprint. 27

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  32. [32]

    Fundamental Limits on Dynamic Inference from Single-Cell Snapshots

    C. Weinreb et al. “Fundamental Limits on Dynamic Inference from Single-Cell Snapshots”.Proceedings of the National Academy of Sciences115.10 (Mar. 2018), E2467–E2476.doi:10.1073/pnas.1714723115

    work page doi:10.1073/pnas.1714723115 2018

  33. [33]

    Electrical Impedance Tomography and Calderón’s Problem

    G. Uhlmann. “Electrical Impedance Tomography and Calderón’s Problem”.Inverse Problems25.12 (Dec. 1, 2009), p. 123011.doi:10.1088/0266-5611/25/12/123011

    work page doi:10.1088/0266-5611/25/12/123011 2009

  34. [34]

    Nonuniqueness in Diffusion-Based Optical Tomography

    S. R. Arridge and W. R. B. Lionheart. “Nonuniqueness in Diffusion-Based Optical Tomography”.Optics Letters23.11 (June 1998), p. 882.doi:10.1364/OL.23.000882

    work page doi:10.1364/ol.23.000882 1998

  35. [35]

    Identifiability of Diffusion Coefficients for Source Terms of Non- Uniform Sign

    M. Bachmayr and V. K. Nguyen. “Identifiability of Diffusion Coefficients for Source Terms of Non- Uniform Sign”.Inverse Problems and Imaging13.5 (2019), pp. 1007–1021.doi:10.3934/ipi.2019045

    work page doi:10.3934/ipi.2019045 2019

  36. [36]

    Parameter Identifiability in PDE Models of Fluorescence Recovery after Photo- bleaching

    M.-V. Ciocanel et al. “Parameter Identifiability in PDE Models of Fluorescence Recovery after Photo- bleaching”.Bulletin of Mathematical Biology86.4 (Apr. 2024), pp. 36–63.doi:10.1007/s11538-024- 01266-4

    work page doi:10.1007/s11538-024- 2024

  37. [37]

    Cox Process Representation and Inference for Stochastic Reaction–Diffusion Processes

    D. Schnoerr, R. Grima, and G. Sanguinetti. “Cox Process Representation and Inference for Stochastic Reaction–Diffusion Processes”.Nature Communications7.1 (May 25, 2016), p. 11729.doi:10.1038/ ncomms11729

    2016

  38. [38]

    Modern Statistics for Spatial Point Processes*

    J. Møller and R. P. Waagepetersen. “Modern Statistics for Spatial Point Processes*”.Scandinavian Journal of Statistics34.4 (Dec. 2007), pp. 643–684.doi:10.1111/j.1467-9469.2007.00569.x

    work page doi:10.1111/j.1467-9469.2007.00569.x 2007

  39. [39]

    Observability and Structural Identifiability of Nonlinear Biological Systems

    A. F. Villaverde. “Observability and Structural Identifiability of Nonlinear Biological Systems”.Com- plexity2019.1 (2019), p. 8497093.doi:10.1155/2019/8497093

    work page doi:10.1155/2019/8497093 2019

  40. [40]

    Effective Drifts in Dynamical Systems with Multiplicative Noise: A Review of Recent Progress

    G. Volpe and J. Wehr. “Effective Drifts in Dynamical Systems with Multiplicative Noise: A Review of Recent Progress”.Reports on Progress in Physics79.5 (May 1, 2016), p. 053901.doi:10.1088/0034- 4885/79/5/053901

    work page doi:10.1088/0034- 2016

  41. [41]

    Langevin Equation in Heterogeneous Landscapes: How to Choose the Interpreta- tion

    A. Pacheco-Pozo et al. “Langevin Equation in Heterogeneous Landscapes: How to Choose the Interpreta- tion”.Physical Review Letters133.6 (Aug. 2024), p. 067102.doi:10.1103/PhysRevLett.133.067102

    work page doi:10.1103/physrevlett.133.067102 2024

  42. [42]

    smiFISH and FISH-quant – a Flexible Single RNA Detection Approach with Super- Resolution Capability

    N. Tsanov et al. “smiFISH and FISH-quant – a Flexible Single RNA Detection Approach with Super- Resolution Capability”.Nucleic Acids Research44.22 (Dec. 15, 2016), e165–e165.doi:10.1093/nar/ gkw784

    work page doi:10.1093/nar/ 2016

  43. [43]

    Chatain et al.Numerical PDE Solvers Outperform Neural PDE Solvers

    P. Chatain et al.Numerical PDE Solvers Outperform Neural PDE Solvers. 2025. arXiv:2507.21269 [math]. Preprint

    work page arXiv 2025

  44. [44]

    Weight Initialization Algorithm for Physics-Informed Neural Networks Using Finite Differences

    H. Tarbiyati and B. Nemati Saray. “Weight Initialization Algorithm for Physics-Informed Neural Networks Using Finite Differences”.Engineering with Computers40.3 (June 2024), pp. 1603–1619.doi: 10.1007/s00366-023-01883-y

    work page doi:10.1007/s00366-023-01883-y 2024

  45. [45]

    Celaya, D

    A. Celaya, D. Fuentes, and B. Riviere.Adaptive Collocation Point Strategies for Physics Informed Neural Networks via the QR Discrete Empirical Interpolation Method. 2025. arXiv:2501.07700 [cs]. Preprint

    work page arXiv 2025

  46. [46]

    Wang et al.An Expert’s Guide to Training Physics-Informed Neural Networks

    S. Wang et al.An Expert’s Guide to Training Physics-Informed Neural Networks. 2023. arXiv:2308. 08468 [cs]. Preprint

    2023

  47. [47]

    Single-Molecule Live-Cell RNA Imaging with CRISPR–Csm

    C. Xia et al. “Single-Molecule Live-Cell RNA Imaging with CRISPR–Csm”.Nature Biotechnology43.12 (Dec. 2025), pp. 2023–2030.doi:10.1038/s41587-024-02540-5

    work page doi:10.1038/s41587-024-02540-5 2025

  48. [48]

    Mechanistic Inference of Stochastic Gene Expression from Structured Single-Cell Data

    C. E. Miles. “Mechanistic Inference of Stochastic Gene Expression from Structured Single-Cell Data”. Current Opinion in Systems Biology42 (Dec. 2025), p. 100555.doi:10.1016/j.coisb.2025.100555

    work page doi:10.1016/j.coisb.2025.100555 2025

  49. [49]

    Using Neural Networks to Estimate Parameters in Spatial Point Process Models

    N. Vihrs. “Using Neural Networks to Estimate Parameters in Spatial Point Process Models”.Spatial Statistics51 (Oct. 2022), p. 100668.doi:10.1016/j.spasta.2022.100668

    work page doi:10.1016/j.spasta.2022.100668 2022

  50. [50]

    Mapping Intracellular Dynamics across the Whole Cell with Spatial Statistics

    Y. Okabe et al. “Mapping Intracellular Dynamics across the Whole Cell with Spatial Statistics”. Biophysical Journal124.23 (Dec. 2025), pp. 4205–4214.doi:10.1016/j.bpj.2025.10.005

    work page doi:10.1016/j.bpj.2025.10.005 2025

  51. [51]

    ELLA: Modeling Subcellular Spatial Variation of Gene Expression within Cells in High-Resolution Spatial Transcriptomics

    J. X. Wang and X. Zhou. “ELLA: Modeling Subcellular Spatial Variation of Gene Expression within Cells in High-Resolution Spatial Transcriptomics”.Nature Communications16.1 (Nov. 2025), p. 9920. doi:10.1038/s41467-025-64867-0. 28

    work page doi:10.1038/s41467-025-64867-0 2025

  52. [52]

    Practical Parameter Identifiability for Spatio-Temporal Models of Cell Invasion

    M. J. Simpson et al. “Practical Parameter Identifiability for Spatio-Temporal Models of Cell Invasion”. Journal of The Royal Society Interface17.164 (Mar. 2020), p. 20200055.doi:10.1098/rsif.2020.0055

    work page doi:10.1098/rsif.2020.0055 2020

  53. [53]

    Identifiability Analysis for Stochastic Differential Equation Models in Systems Biology

    A. P. Browning et al. “Identifiability Analysis for Stochastic Differential Equation Models in Systems Biology”.Journal of The Royal Society Interface17.173 (Dec. 2020), p. 20200652.doi:10.1098/rsif. 2020.0652

    work page doi:10.1098/rsif 2020

  54. [54]

    Diffusion Coefficients Estimation for Elliptic Partial Differential Equations

    A. Bonito et al. “Diffusion Coefficients Estimation for Elliptic Partial Differential Equations”.SIAM Journal on Mathematical Analysis49.2 (Jan. 2017), pp. 1570–1592.doi:10.1137/16M1094476

    work page doi:10.1137/16m1094476 2017

  55. [55]

    Tung and S

    H.-R. Tung and S. D. Lawley.Escape from Heterogeneous Diffusion. Dec. 2025. arXiv:2512.19646 [cond-mat.stat-mech]. Preprint

    work page arXiv 2025

  56. [56]

    Stochastic search with space-dependent diffusivity

    H.-R. Tung and S. D. Lawley. “Stochastic search with space-dependent diffusivity”.Phys. Rev. E113 (6 June 2026), p. 064149.doi:10.1103/cmlr-6yfl

    work page doi:10.1103/cmlr-6yfl 2026

  57. [57]

    Identifying the Interpretation in Two-Dimensional Diffusion Processes with Power- Law Spatially Dependent Diffusivity

    H.-C. Liu et al. “Identifying the Interpretation in Two-Dimensional Diffusion Processes with Power- Law Spatially Dependent Diffusivity”.Physical Review Research8.1 (Feb. 2026), p. 013143.doi: 10.1103/xz7y-n8w2

    work page doi:10.1103/xz7y-n8w2 2026

  58. [58]

    Separable Nonlinear Least Squares: The Variable Projection Method and Its Applications

    G. Golub and V. Pereyra. “Separable Nonlinear Least Squares: The Variable Projection Method and Its Applications”.Inverse Problems19.2 (Apr. 2003), R1–R26.doi:10.1088/0266-5611/19/2/201

    work page doi:10.1088/0266-5611/19/2/201 2003

  59. [59]

    The Immersed Interface Method for Elliptic Equations with Discontinuous Coefficients and Singular Sources

    R. J. LeVeque and Z. Li. “The Immersed Interface Method for Elliptic Equations with Discontinuous Coefficients and Singular Sources”.SIAM Journal on Numerical Analysis31.4 (Aug. 1994), pp. 1019– 1044.doi:10.1137/0731054

    work page doi:10.1137/0731054 1994

  60. [60]

    Solving Inverse Problems in Physics by Optimizing a Discrete Loss: Fast and Accurate Learning without Neural Networks

    P. Karnakov, S. Litvinov, and P. Koumoutsakos. “Solving Inverse Problems in Physics by Optimizing a Discrete Loss: Fast and Accurate Learning without Neural Networks”.PNAS Nexus3.1 (Dec. 2023). Ed. by D. Abbott, pgae005.doi:10.1093/pnasnexus/pgae005

    work page doi:10.1093/pnasnexus/pgae005 2023

  61. [61]

    Physics-Informed Neural Networks: A Deep Learning Framework for Solving Forward and Inverse Problems Involving Nonlinear Partial Differential Equations

    M. Raissi, P. Perdikaris, and G. E. Karniadakis. “Physics-Informed Neural Networks: A Deep Learning Framework for Solving Forward and Inverse Problems Involving Nonlinear Partial Differential Equations”. Journal of Computational Physics378 (Feb. 2019), pp. 686–707.doi:10.1016/j.jcp.2018.10.045

    work page doi:10.1016/j.jcp.2018.10.045 2019

  62. [62]

    A Cusp-Capturing PINN for Elliptic Interface Problems

    Y.-H. Tseng et al. “A Cusp-Capturing PINN for Elliptic Interface Problems”.Journal of Computational Physics491 (Oct. 2023), p. 112359.doi:10.1016/j.jcp.2023.112359

    work page doi:10.1016/j.jcp.2023.112359 2023

  63. [63]

    PyTorch: An Imperative Style, High-Performance Deep Learning Library

    A. Paszke et al. “PyTorch: An Imperative Style, High-Performance Deep Learning Library”.Proceedings of the 33rd International Conference on Neural Information Processing Systems. 721. Red Hook, NY, USA: Curran Associates Inc., Dec. 2019, pp. 8026–8037

    2019

  64. [64]

    BiLO: Bilevel Local Operator Learning for PDE Inverse Problems

    R. Z. Zhang et al. “BiLO: Bilevel Local Operator Learning for PDE Inverse Problems”.Journal of Computational Physics551 (Apr. 2026), p. 114679.doi:10.1016/j.jcp.2026.114679

    work page doi:10.1016/j.jcp.2026.114679 2026

  65. [65]

    R. Z. Zhang et al.Bayesian BiLO: Bilevel Local Operator Learning for Efficient Uncertainty Quantifi- cation of Bayesian PDE Inverse Problems with Low-Rank Adaptation. Jan. 2026. arXiv:2507.17019 [cs]. Preprint

    work page arXiv 2026

  66. [66]

    On the Eigenvector Bias of Fourier Feature Networks: From Regression to Solving Multi-Scale PDEs with Physics-Informed Neural Networks

    S. Wang, H. Wang, and P. Perdikaris. “On the Eigenvector Bias of Fourier Feature Networks: From Regression to Solving Multi-Scale PDEs with Physics-Informed Neural Networks”.Computer Methods in Applied Mechanics and Engineering384 (Oct. 2021), p. 113938.doi:10.1016/j.cma.2021.113938. 29

    work page doi:10.1016/j.cma.2021.113938 2021