IdentityByDescentDispersal.jl: Inferring dispersal rates with identity-by-descent blocks

Arthur Zwaenepoel; Els Lea R De Keyzer; Francisco Campuzano Jim\'enez; Hannes Svardal

REVIEW 2 major objections 4 minor 14 references

IdentityByDescentDispersal.jl: Inferring dispersal rates with identity-by-descent blocks

T0 review · 2 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper presents IdentityByDescentDispersal.jl, a registered Julia package that implements the Ringbauer–Coop–Barton identity-by-descent inference scheme, letting researchers estimate effective population density and per-generation…

desk verdict Useful Julia package implementing RCB17 with a real extension to arbitrary demography, but its single validation example overclaims: the reported 95% CI excludes the true density, so the package's core inference claim needs more evidence. read the letter →

arxiv 2507.06964 v3 pith:LNGJMVUE submitted 2025-07-09 q-bio.PE

Francisco Campuzano Jiménez , Arthur Zwaenepoel , Els Lea R De Keyzer , Hannes Svardal This is my paper · ORCID

classification q-bio.PE

keywords identity-by-descentdispersalrateeffectivepopulationdensitycompositelikelihoodJuliaisolationbydistancegenetics

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

This paper presents IdentityByDescentDispersal.jl, a registered Julia package that turns patterns of shared identity-by-descent (IBD) DNA blocks into estimates of effective population density and per-generation dispersal rate. It fills the stated gap of a general-purpose software implementation of the inference scheme proposed by Ringbauer, Coop, and Barton (2017), which had previously been limited to analytically tractable power-law density histories and gradient-free optimization. The package solves the core integral for the expected density of IBD blocks, evaluates composite likelihoods under a Poisson model, and supports maximum-likelihood and Bayesian inference. It also accepts arbitrary user-defined density functions through numerical integration and provides a simulation template plus an end-to-end pipeline from phased VCF files to estimates. If the implementation is right, researchers studying fragmented or invasive populations can obtain dispersal estimates from spatial genetic data without writing the method from scratch.

What carries the argument

The load-bearing object is Eq. (1), the expected density of IBD blocks: $$\mathbb{E}[N_L \mid r,\$\theta$] = \int_0^\infty G\,$4t^{2}$ $e^{{-2Lt}}$\,\Phi(t\mid r,\$\theta$)\,dt,$$ where $G$ is the genome length in Morgans, $L$ is block length, $r$ is the present-day distance between the two individuals, and $\Phi(t\mid r,\theta)$ is the instantaneous coalescence rate at time $t$ under the demographic model with parameters $\theta$. The package evaluates this integral either analytically for a constant density or a power-law density $D_e(t)=Dt^{-\beta}$, or numerically through Gaussian quadrature for a user-defined $D_e(t)$. It then forms a composite log-likelihood by assuming the number of IBD blocks in a small length bin $[L,L+\Delta L]$ is Poisson with mean $\mathbb{E}[N_L\mid r,\theta]\Delta L$. Automatic differentiation of this composite likelihood is what makes gradient-based optimization and Bayesian sampling practical.

What would settle it

Run many simulated populations with known $D$ and $\sigma$ through the package's own SLiM-to-inference pipeline and measure the empirical coverage of the reported 95% confidence intervals; if the intervals contain the true values less than 95% of the time even when the generative model matches the assumed Poisson/coalescent model, the central inference claim fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the identity-by-descent inference scheme of Ringbauer, Coop, and Barton (2017) can be implemented as a general-purpose, registered Julia package and made practical for ordinary users. The package computes the expected density of IBD blocks per pair of individuals and per unit block length from Eq. (1), forms a composite log-likelihood under the Poisson assumption, and delivers MLEs or Bayesian pseudo-posteriors for the effective population density $D$ and dispersal rate $\sigma$. It extends the original method by allowing arbitrary user-defined density histories $D_e(t)$ through numerical integration and by making the composite likelihood fully compatible with automatic differentiation, including differentiation with respect to the power-law exponent $\beta$. For a simulated constant-density population with true values $D\approx 250$ diploids/km$^2$ and $\sigma\approx 0.071$ km/generation, the worked example reports $D_{\mathrm{MLE}}\approx 282$ diploids/km$^2$ (95% CI 260\textendash 303) and $\sigma_{\mathrm{MLE}}\approx 0.068$ km/generation (95% CI 0.065\textendash 0.071), alongside a Bayesian pseudo-posterior concentrated near the true values.

Load-bearing premise

The whole method rests on the assumption that the mathematical model behind Eq. (1), together with the Poisson composite likelihood, correctly describes how IBD blocks appear in real genomes, so a wrong generative model would bias the estimates even if the code runs perfectly.

Editorial extensions

If this is right

Users can obtain composite-likelihood MLEs or Bayesian posteriors for $D$ and $\sigma$ from IBD data using a registered package rather than reimplementing the underlying equations.
Arbitrary histories of the effective density $D_e(t)$ become estimable numerically, not just the analytically tractable power-law form $Dt^{-\beta}$.
Gradient-based optimization and modern MCMC samplers work directly on the composite likelihood because it is compatible with automatic differentiation, including with respect to $\beta$.
The supplied SLiM template and Snakemake workflow take users from phased VCF files to final estimates, making the method accessible in applied conservation and evolutionary studies.

Reading between the lines

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

A natural next step would be to benchmark the package on real datasets with known demographic histories, such as island populations or invasion fronts, since the paper only demonstrates a simulated example.
Because the expected-block calculation is modular in $\Phi(t\mid r,\theta)$, the same code could be adapted to heterogeneous or anisotropic landscapes without rederiving the integral, although the paper does not propose this.
The paper's own caution that composite-likelihood intervals are overconfident implies that conservation decisions should treat the reported confidence intervals as lower bounds on uncertainty until simulation-based calibration is performed.

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.

Desk Editor's Note

read the letter

You should know one thing up front: this is a genuinely useful software paper, but the validation section is weaker than the contribution. If you need to estimate dispersal from IBD blocks, this package saves you from implementing RCB17 from scratch, and the AD-compatible composite likelihood plus numerical quadrature for arbitrary De(t) are real extensions. The authors are scrupulous about crediting the theory to Ringbauer et al. 2017, and the package is registered, documented, and ships with a SLiM template and a Snakemake pipeline. That is solid engineering and I expect the package to get used.

The soft spot is exactly what the stress-test note flags. There is only one validation dataset: error-free blocks from a constant-density simulation with Dtrue≈250 and σtrue≈0.071. The reported MLE is D≈282 with a 95% CI of 260–303, which excludes the true value, and the CI for σ only just touches its true value. The Figure 1 caption says the pseudo-posterior “is forced to concentrate near the true values,” but the numbers in the text contradict that. The authors even warn that composite-likelihood inference is overconfident, yet they don't discuss why their own interval misses the truth. That mismatch could be an implementation bug, a census-versus-effective density issue, or an unlucky draw amplified by overconfident intervals. The paper doesn't distinguish these.

I don't think this is fatal. The code is transparent, the equations match RCB17, and the extension to custom density models is credible. But the central claim “this implementation recovers D and σ under a known demographic model” is not yet established. A revision with replicate simulations, coverage checks, and some honest discussion of effective-density interpretation would settle it. This is exactly the kind of paper where a serious referee can ask for a small simulation study and the authors can deliver it quickly.

Verdict: send it to peer review. It deserves referee time, not a desk reject, but the referee should push on the validation.

Net: I'd cite this if I were doing IBD-based dispersal inference, and I'd bring it to a methods reading group to discuss the validation standard for software papers. Just don't take the Figure 1 caption at face value until they fix the numbers.

Referee Report

2 major / 4 minor

Summary. The paper introduces IdentityByDescentDispersal.jl, a registered Julia package that implements the identity-by-descent (IBD) block inference scheme of Ringbauer, Coop, and Barton (2017). The package computes the expected density of IBD blocks per pair of individuals via Eq. (1), offers analytic solutions for constant and power-law effective density functions, supports arbitrary user-defined density functions through numerical quadrature, and provides a composite-likelihood objective that is compatible with automatic differentiation, enabling gradient-based maximum-likelihood and Bayesian inference with Turing.jl. The manuscript also describes a SLiM simulation template and a Snakemake pipeline for end-to-end analysis from phased VCF files to parameter estimates. A single validation example on simulated error-free IBD data reports MLEs of D≈282 diploids/km² and σ≈0.068 km/generation, which the authors interpret as success.

Significance. If the package works as claimed, it fills a clear gap: there is no general-purpose, maintained software implementation of the RCB17 method despite its demonstrated usefulness for estimating dispersal from spatial IBD data. The contributions are concrete and partly verifiable: the package is registered, the code is public, the AD-compatible composite likelihood is a genuinely useful engineering advance over the original gradient-free approach, and the extension to arbitrary density functions broadens the method's applicability. The accompanying simulation and bioinformatics pipelines are valuable for adoption. However, the manuscript's only quantitative validation is thin and internally inconsistent, so the central 'it recovers the true parameters' claim is not yet established. The strengths are real, but the evidence for the headline capability needs to be made rigorous before the package can be recommended for routine use.

major comments (2)

[Section 3, last paragraph and Section 4] The paper itself warns in Section 3 that composite-likelihood inference 'tend[s] to be overconfident, underestimating posterior uncertainty and yielding too narrow confidence intervals.' Yet the example uses Fisher-information-based 95% intervals and presents them as supporting the method. The reported interval for D excludes the simulated value, which is exactly the overconfidence failure mode the authors warn about. The manuscript does not address whether the discrepancy is due to the known overconfidence, to a bias in the expected-block density implementation, or to the effective-density interpretation of the SLiM parameters. This needs a concrete treatment: for example, report profile-likelihood intervals, run a small replicate study to show empirical coverage, or at least state explicitly that the intervals are known to be anti-conservative and should not be over-interpreted. As written, the validation does not support the uncertainty quantification that the example advertises.
[Section 4, custom density feature] A headline feature of the package is the ability to fit arbitrary user-defined density functions De(t) through numerical integration (Section 3, Table 1). However, no validation is shown that the numerical quadrature implementation produces results consistent with the analytic solutions for the constant-density and power-density cases. Since the custom-density path is central to the advertised novelty, the authors should add a convergence test or a comparison that demonstrates the numerical integral reproduces the analytic formulas to a stated tolerance over a range of parameter values. Without this, the claim that arbitrary demographic models are 'supported' is not quantitatively supported.

minor comments (4)

[Title and header] The paper title and running header use 'IdentityBydescentDispersal.jl' without a space between 'By' and 'Descent'; this should be corrected to 'IdentityByDescentDispersal.jl' for consistency with the package name and Julia conventions.
[Equation (1)] Equation (1) is typed as 'G4t2 exp(−2Lt)· Φ(t|r,θ )dt', which lacks explicit multiplication signs and superscripts; it should read 'G · 4t² · exp(−2Lt) · Φ(t|r,θ) dt' (or use proper LaTeX). The current rendering is ambiguous and should be fixed for readability.
[Section 3, software description] The text says the package computes 'composite_loglikelihood' functions; the naming is consistent with the code, but the prose would benefit from a brief explanation of why 'composite' likelihood is used rather than full likelihood, especially for readers not familiar with RCB17.
[Section 4, example] The example code snippet uses 'contig_lengths = [1.0]', but the meaning of this value (genome length in Morgans?) is not explained in the text; a one-sentence clarification would help users interpret the pipeline output.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper implements an externally published method, and its simulation example is an independent forward-in-time test rather than a fitted prediction.

full rationale

The paper is a software paper implementing the inference scheme of Ringbauer, Coop, and Barton (2017), an external publication. Its central equations (Eq. 1 for the expected density of IBD blocks and the Poisson composite likelihood) are taken from that external method, not derived from the package's own outputs. The estimates of D and sigma are not baked into the input data by construction: the example simulates IBD blocks with SLiM under fixed true parameters and then estimates those parameters, which is a genuine forward simulation test. The fact that the SLiM template is provided by the authors makes the validation in-family rather than independent, but that is a limitation of the test design, not circular reasoning. The notable mismatch between the reported MLE confidence interval for D (260-303) and Dtrue about 250, and the caption claim that the pseudo-posterior is forced to concentrate near the true values, is a calibration or correctness concern (possibly composite-likelihood overconfidence, effective-density mismatch, an unlucky draw, or an implementation issue), not a circularity. The paper's own warnings about common IBD detection errors and overconfident composite-likelihood inference are limitations, not self-citation load-bearing. No equation reduces to its own input, and no fitted parameter is renamed as a prediction; therefore the circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the previously published RCB17 analytic model, the Poisson composite-likelihood approximation, and standard numerical quadrature. No new biological constants or entities are introduced. The only 'free parameters' are the user-supplied demographic parameters of the inference model, which are not fitted by the paper itself.

assumptions (4)

domain assumption Eq. (1) gives the correct expected density of IBD blocks under the demographic model (from Ringbauer, Coop, and Barton 2017).
The paper adopts RCB17's coalescence-rate formulation without re-derivation; the accuracy of the software is bounded by the accuracy of this model. Invoked in Section 3, Eq. (1).
domain assumption Observed IBD-block counts in discrete length bins follow independent Poisson distributions with mean lambda = E[N_L|r,theta]DeltaL.
Composite likelihood construction in Section 3; the paper itself notes composite likelihoods are overconfident, so this approximation affects uncertainty calibration.
standard math Gauss-Kronrod quadrature converges for the integrals in Eq. (1) over arbitrary user-defined De(t).
Numerical integration is assumed accurate; for rough or discontinuous density functions the user must verify convergence. Mentioned in Section 2 via QuadGK.
domain assumption SLiM forward simulation recapitulates the same spatial coalescent process.
Used in Section 4 to generate the validation dataset; if SLiM and the analytic model diverge, the demo's recovery of true parameters understates model error.

how reviews work

0 comments

Cite this review

Pith. "Pith review of IdentityByDescentDispersal.jl: Inferring dispersal rates with identity-by-descent blocks." pith.science (2026). https://pith.science/paper/LNGJMVUE

@misc{pith2026250706964,
  author       = {Pith},
  title        = {Pith review of: IdentityByDescentDispersal.jl: Inferring dispersal rates with identity-by-descent blocks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LNGJMVUE}},
  note         = {Machine review of arXiv:2507.06964}
}

read the original abstract

The population density and per-generation dispersal rate of a population are central parameters in the study of evolution and ecology. The distribution of recent coalescent events between individuals in space can be used to estimate such quantities through the distribution of identity-by-descent (IBD) blocks. An IBD block is defined as a segment of DNA that has been inherited by a pair of individuals from a common ancestor without being broken by recombination. We introduce a Julia package for estimating effective population densities and dispersal rates from observed spatial patterns of IBD shared blocks. It implements the inference scheme proposed by Ringbauer, Coop, and Barton (2017). The package provides a user-friendly interface, supports efficient gradient-based optimization and accommodates arbitrary user-defined demographic models through numerical integration. This software aims to encourage a wider audience to utilize spatial genetic data for estimating dispersal rates, thereby motivating further research and expanding its applications.

Figures

Figures reproduced from arXiv: 2507.06964 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 11 canonical work pages

[8]

QuadGK.jl: Gauss–Kronrod Integration in Julia

“QuadGK.jl: Gauss–Kronrod Integration in Julia.” https://github.com/JuliaMath/QuadGK.jl. 4 IDENTITY BYDESCENT DISPERSAL .JL - AUGUST 21, 2025 Mölder, Felix, Kim Philipp Jablonski, Brice Letcher, Michael B. Hall, Christopher H. Tomkins-Tinch, Vanessa Sochat, Jan Forster, et al

work page 2025
[10]

Inferring Recent Demography from Isolation by Distance of Long Shared Sequence Blocks

“Inferring Recent Demography from Isolation by Distance of Long Shared Sequence Blocks.” Genetics 205 (3): 1335–51. https://doi.org/10.1534/genetics.116.196

work page doi:10.1534/genetics.116.196
[12]

Dispersal Inference from Population Genetic Variation Using a Convolutional Neural Network

“Dispersal Inference from Population Genetic Variation Using a Convolutional Neural Network.”Genetics 224 (2): iyad068. https://doi.org/10.1093/genetics/iya d068. Smith, Chris C. R., and Andrew D. Kern

work page doi:10.1093/genetics/iya
[33]

Ringbauer, Harald, Graham Coop, and Nicholas H

https: //doi.org/10.12688/f1000research.29032.2. Ringbauer, Harald, Graham Coop, and Nicholas H. Barton

doi:10.12688/f1000research.29032.2
[385]

Zhou, Ying, Sharon R

https://doi.org/10.1186/s12859-023- 05522-7. Zhou, Ying, Sharon R. Browning, and Brian L. Browning

work page doi:10.1186/s12859-023-
[1997]

Genetic Differentiation and Estimation of Gene Flow from F-Statistics Under Isolation by Distance

“Genetic Differentiation and Estimation of Gene Flow from F-Statistics Under Isolation by Distance.” Genetics 145 (4): 1219–28. https://doi.org/10.1093/genetics/145.4.1219. Smith, Chris C R, Silas Tittes, Peter L Ralph, and Andrew D Kern

work page doi:10.1093/genetics/145.4.1219
[2012]

Identity by Descent Between Distant Relatives: Detection and Applications

“Identity by Descent Between Distant Relatives: Detection and Applications.” Annual Review of Genetics 46: 617–33. https://doi.org/10.1146/annurev-genet-110711-155534. Driscoll, Don A., Sam C. Banks, Philip S. Barton, Karen Ikin, Pia Lentini, David B. Lindenmayer, Annabel L. Smith, et al

work page doi:10.1146/annurev-genet-110711-155534
[2013]

Improving the Accuracy and Efficiency of Identity-by-Descent Detection in Population Data

“Improving the Accuracy and Efficiency of Identity-by-Descent Detection in Population Data.” Genetics 194 (2): 459–71. https://doi.org/10.1534/genetics.113.150029. Browning, Sharon R., and Brian L. Browning

work page doi:10.1534/genetics.113.150029

Show all 14 references

[2014]

The Trajectory of Dispersal Research in Conservation Biology. Systematic Review

“The Trajectory of Dispersal Research in Conservation Biology. Systematic Review.” PLOS ONE 9 (4): e95053. https://doi.org/10.1371/journal.pone.0095053. Ge, Hong, Kai Xu, and Zoubin Ghahramani

doi
[2017]

Julia: A Fresh Approach to Numerical Computing

“Julia: A Fresh Approach to Numerical Computing.” SIAM Review 59 (1): 65–98. https://doi.org/10.1137/141000671. Browning, Brian L., and Sharon R. Browning

doi
[2019]

Tree-Sequence Recording in SLiM Opens New Horizons for Forward-Time Simulation of Whole Genomes

“Tree-Sequence Recording in SLiM Opens New Horizons for Forward-Time Simulation of Whole Genomes.” Molecular Ecology Resources 19 (2): 552–66. https://doi.org/10.1111/1755-0998.12968. Haller, Benjamin C., and Philipp W. Messer
[2020]

A Fast and Simple Method for Detecting Identity- by-Descent Segments in Large-Scale Data

“A Fast and Simple Method for Detecting Identity- by-Descent Segments in Large-Scale Data.” American Journal of Human Genetics 106 (4): 426–37. https: //doi.org/10.1016/j.ajhg.2020.02.010. 5

2020 doi
[2022]

Fitting Matérn Smoothness Parameters Using Automatic Differentiation

“Fitting Matérn Smoothness Parameters Using Automatic Differentiation.” arXiv. https://doi.org/10.48550/ARXIV.2201.00090. Haller, Benjamin C., Jared Galloway, Jerome Kelleher, Philipp W. Messer, and Peter L. Ralph

doi arXiv
[2023]

SLiM 4: Multispecies Eco-Evolutionary Modeling

“SLiM 4: Multispecies Eco-Evolutionary Modeling.” The American Naturalist 201 (5): E127–39. https://doi.org/10.1086/723601. Johnson, Steven G

doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [8]

QuadGK.jl: Gauss–Kronrod Integration in Julia

“QuadGK.jl: Gauss–Kronrod Integration in Julia.” https://github.com/JuliaMath/QuadGK.jl. 4 IDENTITY BYDESCENT DISPERSAL .JL - AUGUST 21, 2025 Mölder, Felix, Kim Philipp Jablonski, Brice Letcher, Michael B. Hall, Christopher H. Tomkins-Tinch, Vanessa Sochat, Jan Forster, et al

work page 2025

[2] [10]

Inferring Recent Demography from Isolation by Distance of Long Shared Sequence Blocks

“Inferring Recent Demography from Isolation by Distance of Long Shared Sequence Blocks.” Genetics 205 (3): 1335–51. https://doi.org/10.1534/genetics.116.196

work page doi:10.1534/genetics.116.196

[3] [12]

Dispersal Inference from Population Genetic Variation Using a Convolutional Neural Network

“Dispersal Inference from Population Genetic Variation Using a Convolutional Neural Network.”Genetics 224 (2): iyad068. https://doi.org/10.1093/genetics/iya d068. Smith, Chris C. R., and Andrew D. Kern

work page doi:10.1093/genetics/iya

[4] [33]

Ringbauer, Harald, Graham Coop, and Nicholas H

https: //doi.org/10.12688/f1000research.29032.2. Ringbauer, Harald, Graham Coop, and Nicholas H. Barton

doi:10.12688/f1000research.29032.2

[5] [385]

Zhou, Ying, Sharon R

https://doi.org/10.1186/s12859-023- 05522-7. Zhou, Ying, Sharon R. Browning, and Brian L. Browning

work page doi:10.1186/s12859-023-

[6] [1997]

Genetic Differentiation and Estimation of Gene Flow from F-Statistics Under Isolation by Distance

“Genetic Differentiation and Estimation of Gene Flow from F-Statistics Under Isolation by Distance.” Genetics 145 (4): 1219–28. https://doi.org/10.1093/genetics/145.4.1219. Smith, Chris C R, Silas Tittes, Peter L Ralph, and Andrew D Kern

work page doi:10.1093/genetics/145.4.1219

[7] [2012]

Identity by Descent Between Distant Relatives: Detection and Applications

“Identity by Descent Between Distant Relatives: Detection and Applications.” Annual Review of Genetics 46: 617–33. https://doi.org/10.1146/annurev-genet-110711-155534. Driscoll, Don A., Sam C. Banks, Philip S. Barton, Karen Ikin, Pia Lentini, David B. Lindenmayer, Annabel L. Smith, et al

work page doi:10.1146/annurev-genet-110711-155534

[8] [2013]

Improving the Accuracy and Efficiency of Identity-by-Descent Detection in Population Data

“Improving the Accuracy and Efficiency of Identity-by-Descent Detection in Population Data.” Genetics 194 (2): 459–71. https://doi.org/10.1534/genetics.113.150029. Browning, Sharon R., and Brian L. Browning

work page doi:10.1534/genetics.113.150029