Local ancestry inference identifies robust evidence of selection in Neolithic Europe

Name: Verified Rigor Record — Local ancestry inference identifies robust evidence of selection in Neolithic Europe
Item: Local ancestry inference identifies robust evidence of selection in Neolithic Europe
Rating: 94
Author: Αlpha¹

Mies, G.; Mathieson, I.

Paper source

Read PDF

Αlpha¹ Rigor Review

Rigor Review Verified

Critical

High-priority

Action items to resolve · Critical / High

Confidence: highEngine: 7.6.0Source: full text

Not verified in this report

References were not verified against Crossref/OpenAlex.

This Alpha1 Rigor Review uses AI Rigor Reviewers trained on a curated corpus of high-fidelity and retracted papers, with expert supervision and curation. It can still make mistakes; verify each finding against the source before relying on it.

The paper is a well-structured computational benchmarking study with strong replication and data sharing. The main weaknesses are incomplete reporting of statistical details (exact p-values, effect sizes with confidence intervals, assumptions verification) and missing version numbers for some software tools, which reduce reproducibility. The design is otherwise sound, and the conclusions are appropriately cautious.

Evaluated based on full text of the preprint. Three independent reviewers scored all eight dimensions; their evidence was synthesized. The statistics component recomputed 3 tests consistently, but this does not verify all reported statistics. The citation and claim audit components raised no concerns. The copyedit pass identified minor clarity and punctuation issues.

9 major claims checked against the paper's own evidence: all adequately supported.

supportedResults, paragraph 5Reviewer 1
Integrating results across LAI methods identifies robust ancestry deviations at SLC24A5 and FADS1/2, consistent with adaptation on pigmentation and metabolism.
The paper provides direct evidence: these loci exceed Z > 3 in the discovery meta-analysis, replicate in two independent datasets (P < 0.005 and P < 0.05), and show decreased deviation in older individuals with low HG ancestry, consistent with selection acting over time.
Evidence: Results paragraph 5: 'Of the ten regions exceeding the discovery threshold, five replicated within 2 Mb of the lead SNP at a significance threshold of P < 0.005 in replication dataset 1, and two of these also replicated in replication dataset 2. The two regions replicated in both replication datasets correspond to... SLC24A5 on chromosome 15... FADS1/2 locus on chromosome 11.' Figure 3 and Supplementary Figures S18-S20 also support.
“Of the ten regions exceeding the discovery threshold, five replicated within 2 Mb of the lead SNP at a significance threshold of P < 0.005 in replication dataset 1, and two of these also replicated in replication dataset 2.”
supportedResults, paragraphs 5-6Reviewer 1
PER3 and IRAK4 are candidate loci for selection but with less consistent signals across methods.
The paper presents evidence that these loci exceed the Z > 3 threshold and replicate in dataset 1 but not dataset 2, and shows elevated deviation in low-HG individuals, consistent with the stated caveat that signals are less robust.
Evidence: Results paragraph 5: 'A fourth signal on chromosome 1 spans multiple genes; a plausible candidate in this region is PER3... The fifth replicated region on chromosome 12 also overlaps multiple genes, with IRAK4... representing a possible candidate.' Results paragraph 6: 'Deviation at PER3 and IRAK4 is consistent across methods but high in the low-ancestry individuals, suggesting that these might reflect ancestry misclassification.'
“Deviation at PER3 and IRAK4 is consistent across methods but high in the low-ancestry individuals, suggesting that these might reflect ancestry misclassification rather than the effects of selection.”
supportedResults, paragraph 6; Discussion, paragraph 3Reviewer 1
Previous reports of excess hunter-gatherer ancestry at the HLA are replicated but may be affected by bias in local ancestry inference.
The paper shows the HLA region exceeds the Z > 3 threshold in discovery and replicates in dataset 1, but finds it inconsistent across methods, with high deviation in low-HG individuals, supporting the claim that the signal may be artifactual.
Evidence: Results paragraph 6: 'Deviation at the HLA is inconsistent across methods and replication datasets and high in the low-ancestry individuals. In summary, ancestry deviations at SLC24A5 and FADS1/2 likely indicate selection at those loci while other deviations, particularly at the HLA, likely indicate that ancestry is being systematically misclassified.' Discussion paragraph 3: 'We conclude that local ancestry at the HLA may be too unreliable to make reliable inference about natural selection or trait associations.'
“Deviation at the HLA is inconsistent across methods and replication datasets and high in the low-ancestry individuals, suggesting that ancestry is being systematically misclassified at those loci.”
supportedResults, paragraphs 3-4Reviewer 1
LAI methods differ by over an order of magnitude in inferred tract lengths, with RFMix producing the most biologically plausible results.
The paper directly compares tract length distributions across methods and shows quantitative differences (e.g., TRACTS estimates: 10.3 generations for RFMix vs. 41 for simpLAI vs. 6.5 for Mosaic), and uses theoretical expectations to judge plausibility.
Evidence: Results paragraph 3: 'Tract length distributions reflect inferred time since admixture... We observed that LAI methods differed by over an order of magnitude in inferred tract lengths: Ancestry HMM, AncestralPaths, and simpLAI produced shorter average tracts compared to RFMix and Mosaic.' Results paragraph 4: 'Using TRACTS, the best-fit admixture timing estimates were 10.3, 41, and 6.5 generations for RFMix, simpLAI, and Mosaic, respectively. Overall, we find that RFMix produces the most biologically plausible tract length distributions.'
“We observed that LAI methods differed by over an order of magnitude in inferred tract lengths: Ancestry HMM, AncestralPaths, and simpLAI produced shorter average tracts compared to RFMix and Mosaic.”
supportedResults, paragraph 2; Discussion, paragraph 1Reviewer 1
Method choice critically influences inferred ancestry patterns and selection signals, underscoring the importance of multi-method validation.
This is the overarching conclusion and is supported by the systematic comparisons showing low local ancestry correlations (0.08-0.5), divergent tract lengths, and method-dependent selection signals.
Evidence: Results paragraph 2: 'Pearson correlations of hunter-gatherer ancestry proportions in 100 kb bins between pairs of methods ranged from 0.08 to 0.5.' Discussion paragraph 1: 'By systematically evaluating multiple LAI approaches, we show that existing methods are highly sensitive to source composition and technical aspects of data preparation.'
“Pearson correlations of hunter-gatherer ancestry proportions in 100 kb bins between pairs of methods ranged from 0.08 to 0.5.”
supportedResults, paragraph 1-2Reviewer 2
Local ancestry inference methods produce correlated but not identical results in ancient DNA, and multi-method integration improves robustness.
The paper presents clear evidence of correlations (e.g., pairwise correlations of 0.08-0.5) and tract length discrepancies, supporting the need for multi-method integration.
Evidence: Results section 'Meta-analysis identifies five regions', correlations in 100kb bins, tract length analysis.
“While individual-level ancestry estimates are highly correlated across methods, inferred tract lengths and admixture time estimates vary by over an order of magnitude.”
supportedResults, Meta-analysis identifies five regionsReviewer 2
SLC24A5 and FADS1/2 show robust ancestry deviations consistent with selection.
The evidence is strong: replication in two datasets, consistency across methods, and temporal pattern.
Evidence: The paper shows that these two loci replicate in both independent datasets (P < 0.005), have directionally consistent Z-scores across methods, and are lower in older, low-admixture individuals.
“The two regions replicated in both replication datasets correspond to previously reported ancestry outliers... SLC24A5 on chromosome 15, consistent with immune-related and pigmentation-associated selection, respectively.”
supportedResults, paragraph 8Reviewer 2
PER3 and IRAK4 are candidate selected loci but with weaker evidence.
The claim is appropriately cautious and supported by the data presented.
Evidence: The paper explicitly states these signals are less consistent and may reflect misclassification, which is accurate.
“Deviation at PER3 and IRAK4 is consistent across methods but high in the low-ancestry individuals, suggesting that these might reflect ancestry misclassification rather than the effects of selection.”
supportedDiscussionReviewer 2
HLA ancestry deviation may be an artifact of LAI bias.
The claim is supported by the presented analysis.
Evidence: The paper shows the HLA signal is inconsistent across methods, not replicated in both datasets, and appears in low-ancestry individuals.
“We conclude that local ancestry at the HLA may be too unreliable to make reliable inference about natural selection or trait associations.”

Recomputed 3 tests: 3 consistent, 0 inconsistent, 3 via agent-written checks.

2 integrity concerns flagged (0 high).

Rigor

Eight Dimensions5 pass · 1 warn

9 major claims9 supported

13 copyedit issues flagged (1 major): mostly clarity, punctuation, typo.

Reproducibility

6 data/code links checked; 6 live.

datahttps://erda.ku.dk/archives/917f1ac64148c3800ab7baa29402d088/published-archive.htmlLIVEHTTP 200
Resolves, but the content could not be matched to the paper.
dataDataverseLIVEHTTP 202
https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/7RVV9N
Resolves to Dataverse (data repository).
dataZenodoLIVEHTTP 200
https://zenodo.org/records/19684580
Resolves to Zenodo (data repository).
codeGitHubLIVEHTTP 200
https://github.com/gmies/LAI_neolithic/
Resolves to GitHub (code repository).
codeGitHubLIVEHTTP 200
https://github.com/CMPG/simpLAI/blob/main/README.md
Resolves to GitHub (code repository).
codeGitHubLIVEHTTP 200
https://github.com/Atkinson-Lab/Tractor
Resolves to GitHub (code repository).

Alpha¹ Rigor Action ItemsFinal pass before journal submission

The paper is robust for a post-publication audit; the main conclusions are supported by the data. An informed reader should weigh the sensitivity of results to method choice, which the paper discusses, and note that the statistical reporting gaps (absence of exact p-values and effect sizes for many comparisons) limit the ability to assess the strength of evidence. An erratum addressing these reporting gaps and adding software version numbers would strengthen the paper.

1.
HIGHstatistics
Report exact p-values (e.g., P = 0.003) for all key statistical tests, including correlation coefficients between LAI methods and comparisons with qpAdm/ADMIXTURE, instead of reporting only thresholds such as P < 0.005.
Exact p-values allow readers to assess the strength of evidence precisely rather than relying on arbitrary thresholds.
2.
HIGHstatistics
Include effect sizes (e.g., Pearson r, regression slopes, mean ancestry deviation) and their 95% confidence intervals for all pairwise method comparisons and for the key selection signals.
Effect sizes with confidence intervals quantify the magnitude and uncertainty of differences, which is essential for interpreting the practical significance of the results.
3.
HIGHstatistics
State whether statistical assumptions (e.g., normality for t-tests, linearity for regression) were verified, and if so, describe the methods used; alternatively, note that robust methods were employed.
Unverified assumptions can undermine the validity of the statistical tests, and readers need this information to judge the reliability of the inferences.
4.
HIGHdata code
Add explicit version numbers and URLs for all software tools used, including Mosaic, SimpLAI, and Ancestry HMM, and provide the parameters and version of the AncestralPaths method (which is only cited as 'published results').
Full reproducibility requires unambiguous identification of software versions; missing version numbers prevent others from replicating the exact computational environment.
5.
MEDIUMreporting
Clarify the inclusion/exclusion criteria for individuals in a more formal way, such as a supplementary table listing all included and excluded samples with reasons, and reference a pre-registered protocol if available.
Even though criteria are described, a structured table improves transparency and allows readers to assess potential selection biases.
6.
MEDIUMreporting
Define the term 'replication' more precisely in the Methods: specify whether the same method and thresholds were used, or if different methods were applied across datasets, and state the criteria for a successful replication.
Ambiguity in what constitutes replication can confuse readers and weaken the interpretation of the validation results.
7.
MEDIUMreporting
Add a brief statement in the Methods that the study reuses only publicly available, de-identified ancient DNA data, so ethics approval was not required, and that the original studies provided appropriate ethics documentation.
A clear statement preempts reader concern about ethical oversight of human remains data, even when reuse is permissible.
8.
MEDIUMreporting
Add a note about whether any reporting guideline (e.g., STREGA for genetic association studies) was followed, or explicitly state that no applicable guideline exists for this study type.
Journals increasingly expect authors to mention reporting guidelines; a brief note would strengthen the transparency section.
9.
MEDIUMcopyedit
Revise the sentence in Results paragraph 2 to clarify that correlations were computed between pairs of LAI methods from hunter-gatherer ancestry proportions in 100 kb bins (copyedit major issue).
The original phrasing is ambiguous about which correlations are being compared, which could lead to misinterpretation.
10.
MEDIUMcopyedit
Provide a supplementary table listing all software versions, URLs, and parameters (including for AncestralPaths) to ensure full reproducibility.
This consolidates the information that is currently scattered across the Methods and improves reproducibility.
11.
LOWcopyedit
Correct minor punctuation and typographical errors: remove extra space before period after the GitHub URL in the Data and code availability section, and change 'useable' to 'usable' in Results paragraph 1.
These minor issues do not affect comprehension but should be corrected for a polished manuscript.
12.
LOWcopyedit
Rephrase the transition from the problem statement to the approach in the Abstract (e.g., 'remains unresolved; we benchmark...' or 'Here, we benchmark...') to improve flow.
The current abrupt transition slightly reduces readability.

Full text (39,005 chars)Analyzed: Jul 7, 2026

Αlpha¹ assesses methodological rigor, not the importance of the findings. See how we calibrate →