SweepLink: Joint Inference of Demography and Linked~Selection from Time-series Data
- 1University of Edinburgh, UK
Abstract
Genome-wide time-series data, i.e. allele frequency trajectories tracked across multiple sampling times, are among the richest sources of information for inferring selection. Beyond a beneficial allele's own rise in frequency, such data capture how it drags nearby loci upward via linkage, an effect known as genetic hitch-hiking. Yet most existing tools are single-locus, treating loci independently: they infer site-specific selection coefficients in isolation, then rely on ad hoc window statistics to account for hitch-hiking. Many existing tools further require a predefined population size, or scale poorly when jointly inferring selection and demography, and their power is highly sensitive to a significance threshold. To address these shortcomings, we here present SweepLink, a two-layer Hidden Markov Model that overcomes these limitations by jointly inferring demography and linked selection genome-wide: a spatial layer captures correlations between neighboring selection coefficients, coupled with a temporal Wright-Fisher diffusion layer. As we show with extensive simulations, this setup pushes drift-driven false signals toward neutrality while reinforcing loci that receive support from neighbouring loci, thereby increasing the sensitivity for weak and moderate selection, while matching the power of existing tools to detect strong selection. These simulations further show that SweepLink yields confident posteriors that remain stable at maximal significance, removing the need for arbitrary thresholds. We applied SweepLink to ancient DNA time-series data from the British population, previously analysed with a single-locus tool. SweepLink recovers four of the previously reported signals (LCT, SLC45A2, DHCR7, HERC2), and partially recovers the MHC/HLA signal. It also identifies additional candidate regions, including DPYD, FADS1/2 and OAS1, missed by the prior scan but supported by independent studies.
Related
- Demography and selection shape gene loss in freshwater sticklebacks - A pangenomic perspective
- Detecting Selection from Linked Sites Using an F-Model
- Stronger positive and purifying selection on X-linked genes in a species with paternal genome elimination
- Thinking too positive? Revisiting current methods of population genetic selection inference
- Explainable AI-driven marker selection and development of EastFocus AISNPplex 61 assay for fine-scale biogeographical ancestry inference
BibTeX
@misc{noskova2026sweeplink,
title = {{SweepLink: Joint Inference of Demography and Linked~Selection from Time-series Data}},
author = {E. Noskova and M. Caduff and A. Fueglistaler and A. Parker and C. Leuenberger and D. Wegmann},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.02.748944},
url = {https://www.biorxiv.org/content/10.64898/2026.09.02.748944v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - SweepLink: Joint Inference of Demography and Linked~Selection from Time-series Data
AU - E. Noskova
AU - M. Caduff
AU - A. Fueglistaler
AU - A. Parker
AU - C. Leuenberger
AU - D. Wegmann
PY - 2026
DA - 2026/09/02
PB - bioRxiv
DO - 10.64898/2026.09.02.748944
UR - https://www.biorxiv.org/content/10.64898/2026.09.02.748944v1
AB - Genome-wide time-series data, i.e. allele frequency trajectories tracked across multiple sampling times, are among the richest sources of information for inferring selection. Beyond a beneficial allele's own rise in frequency, such data capture how it drags nearby loci upward via linkage, an effect known as genetic hitch-hiking. Yet most existing tools are single-locus, treating loci independently: they infer site-specific selection coefficients in isolation, then rely on ad hoc window statistics to account for hitch-hiking. Many existing tools further require a predefined population size, or scale poorly when jointly inferring selection and demography, and their power is highly sensitive to a significance threshold. To address these shortcomings, we here present SweepLink, a two-layer Hidden Markov Model that overcomes these limitations by jointly inferring demography and linked selection genome-wide: a spatial layer captures correlations between neighboring selection coefficients, coupled with a temporal Wright-Fisher diffusion layer. As we show with extensive simulations, this setup pushes drift-driven false signals toward neutrality while reinforcing loci that receive support from neighbouring loci, thereby increasing the sensitivity for weak and moderate selection, while matching the power of existing tools to detect strong selection. These simulations further show that SweepLink yields confident posteriors that remain stable at maximal significance, removing the need for arbitrary thresholds. We applied SweepLink to ancient DNA time-series data from the British population, previously analysed with a single-locus tool. SweepLink recovers four of the previously reported signals (LCT, SLC45A2, DHCR7, HERC2), and partially recovers the MHC/HLA signal. It also identifies additional candidate regions, including DPYD, FADS1/2 and OAS1, missed by the prior scan but supported by independent studies.
ER -