Review Papers Claim by Claim

1) What they did (short)

• Admixture-aware per-variant likelihood scan across 1,150,640 1240k sites in five continental regions using inferred ADMIXTURE proportions, with genomic-control and LD filters to reduce technical false positives Methods (admixture-aware scan)
• Developed bmws-admix: a particle-filter/Gibbs sampler time-series method that models selection in ancestry components and regularizes temporal and cross-population similarity (hyperpriors on α, β) to estimate s(t) trajectories in admixed populations bmws-admix time-series
• Polygenic test: counts trait-increasing alleles across GWAS hits (effect direction only), tests against admixture-conditioned expectation and obtains permuted P-values, excluding GWAS discovery population when testing (to reduce stratification confounding) Polygenic-method summary

2) Core results (evidence-linked)

• 31 genome-wide significant loci reported overall: Europe 24, East Asia 4, Americas 3; none reached genome-wide significance in Africa or South Asia (authors attribute this to lower power) 31 selection peaks summary
• Known targets recovered (LCT, ADH1B, FADS1, pigmentation loci, multiple HLA hits) and plausible novel loci; several loci show cross-region sharing (e.g., ADH1B and FADS1 in Europe & East Asia), interpreted as parallel selection, selection on shared standing variation, or selection prior to divergence Loci recovered & sharing
• Time-series inference shows temporally varying s: LCT in Europe rose strongly until ~100 generations ago then weakened (consistent with prior Bronze-Age work), ADH1B shows intensifying selection in East Asia beginning ~100–150 generations ago; FADS1 shows regional differences in trajectory and magnitude Time-series examples
• Polygenic signals: strong stabilizing selection on female waist-hip ratio (WHR) across populations, directional or stabilizing signals for T2D and immune traits (e.g., rheumatoid arthritis) identified when using non-discovery GWAS to reduce stratification bias Polygenic trait findings

3) Strengths (concise, evidence-linked)

1. Global scope: extends ancient-DNA selection work beyond Europe using curated AADR + present-day panels, increasing comparative power to detect shared vs local adaptation Sample composition and scope
2. Methodological care: admixture-aware null, LD and neighbor-filtering, separate HLA calibration, genomic control, and time-series regularization reduce common confounders in ancient selection scans Method robustness steps
3. Transparent reproducibility: summary statistics/selection scores on Zenodo and code repositories (eas_selection, bmws_admix) publicly listed for reanalysis Data & code availability

4) Limitations, risks, and potential biases (critical)

• Power imbalance: simulations and results both show much higher power in Europe (sample-rich) than Africa or South Asia; absence of hits in Africa/South Asia is likely power-limited, not proof of no selection Power limitation statement
• 1240k ascertainment & low-coverage aDNA: analyses restricted to 1240k sites bias toward common/known variants; pseudohaploid ancient genotypes and dataset merging risk genotype errors and technical artifacts despite filtering (authors filtered variants flagged as potential technical artifacts) 1240k and aDNA caveats
• GWAS stratification risk in polygenic tests: authors mitigate by excluding GWAS discovery population when testing, but residual confounding from differences in LD, effect-size portability, and polygenic architecture across ancestries remains a real risk and could inflate or deflate polygenic signals Polygenic-scan cautions (review)
• HLA complexity: authors used bespoke calibration, but HLA’s extreme LD, balancing selection history, and variable SNP density make localization and haplotype interpretation uncertain — repeated signals may reflect different causal haplotypes or artifacts of mapping/LD HLA calibration and caution
• Model assumptions in bmws-admix: particle filters and regularizers require choices (P particles, α/β) and assume known/constant N_e in simplest implementation; authors set α,β after cross-validation but time-varying N_e, unmodeled admixture pulses, or mis-specified K could bias s(t) estimates (authors acknowledge these modeling assumptions) bmws-admix modeling assumptions

5) Specific reproducibility checklist (practical)

1. Download selection scores & P-values from Zenodo DOI provided in paper and confirm Manhattan peak coordinates match Supplementary Table 1 Zenodo data link
2. Obtain the same 1240k filtered genotype matrix from the AADR and the 1kG pseudohaploid projection used by the authors to replicate ADMIXTURE M matrix inference (matching K per-region) Population panel & ADMIXTURE
3. Run a small locus-level replication: compute M for chosen region, fit Q per SNP, compute likelihood ratio and genomic-control, and compare to reported P for e.g. FADS1, ADH1B, LCT (authors supply LocusZoom plots and data) Locus replication recommendation

6) Confidence, open questions, and what would overturn key claims

Confidence in major claims:

• Identification of canonical loci (LCT, FADS1, ADH1B, pigmentation loci): high confidence — these loci are repeatedly recovered in independent ancient-DNA studies and match prior expectations Canonical loci detection
• Cross-regional sharing interpretation: moderate confidence — shared signals are real, but their interpretation (parallel selection vs shared ancestry vs recurrent mutation) needs fine-mapping and haplotype-resolution to decide; time-series/haplotype work in richer datasets would clarify mechanisms Shared-signal interpretation

What would overturn main conclusions:

• Independent re-analyses with larger non-European ancient sample sets (Africa, South Asia) that find similar numbers of hits outside Europe would change the apparent Europe-centric distribution — conversely, more data showing persistent nulls would strengthen claims of Europe-predominant recent sweeps.
• Demonstration that polygenic signals (WHR, T2D, RA) vanish when using robust cross-ancestry GWAS and within-ancestry effect-size corrections would overturn claims about universal stabilizing selection on WHR.

7) Practical suggestions to improve/extend the study (actionable)

1. Add dense haplotype-based analyses (e.g., Relate/ARG-based methods or imputation + local-ancestry-aware haplotype tests) at shared loci to determine whether the same haplotype swept independently or distinct alleles were targeted.
2. Increase targeted ancient sampling in underpowered regions (Africa, South Asia) prioritizing stratified temporal transects and higher-coverage genomes; prioritize imputation-capable sites (or shotgun sequencing) beyond 1240k to reduce ascertainment bias.
3. For polygenic tests: repeat using multi-ancestry GWAS or within-ancestry effect estimates (or leave-one-region-out meta-GWAS) and quantify sensitivity of signals (WHR) to effect-size portability and LD differences.