Timesweeper: Accurately Identifying Selective Sweeps Using Population Genomic Time Series

Abstract

Despite decades of research, identifying selective sweeps, the genomic footprints of positive selection, remains a core problem in population genetics. Of the myriad methods that have been made towards this goal, few are designed to leverage the potential of genomic time-series data. This is because in most population genetic studies only a single, brief period of time can be sampled for a study. Recent advancements in sequencing technology, including improvements in extracting and sequencing ancient DNA, have made repeated samplings of a population possible, allowing for more direct analysis of recent evolutionary dynamics. With these advances in mind, here we present Timesweeper, a fast and accurate convolutional neural network-based tool for identifying selective sweeps in data consisting of multiple genomic samplings of a population over time. Timesweeper utilizes this serialized sampling of a population by first simulating data under a demographic model appropriate for the data of interest, training a 1D Convolutional Neural Network on said model, and inferring which polymorphisms in this serialized dataset were the direct target of a completed or ongoing selective sweep. We show that Timesweeper is accurate under multiple simulated demographic and sampling scenarios, and identifies selected variants with impressive resolution. In sum, we show that more accurate inferences about natural selection are possible when genomic time-series data are available; such data will continue to proliferate in coming years due to both the sequencing of ancient samples and repeated samplings of extant populations with faster generation times. Methodological advances such as Timesweeper thus have the potential to help resolve the controversy over the role of positive selection in the genome. We provide Timesweeper both as a Python package and Snakemake workflow for use by the community.

Deep Learning in Population Genetics

“…One key aspect to make deep learning a popular framework in population genetics, is to ensure reproducible analyses and avoid repeating training of highly parameterized networks from scratch. In this context, recent efforts to provide users with documented workflows ( Whitehouse and Schrider 2022 ) and pre-trained networks ( Hamid et al 2022 ) will both reduce carbon footprint ( Grealey et al 2022 ) and facilitate the application of deep learning to a wider range of data sets, allowing users to modify the network’s parameters according to the specific requirements of the biological system under examination.…”

“…In recent years, the generation of sequencing data from ancient or historical samples, as well as from capture-recapture and evolve-and-resequence experiments, has allowed for a direct observation of how genetic diversity and allele frequencies change under natural or controlled conditions over time. To detect positive selection with time-series data, Whitehouse and Schrider (2022) proposed to stack either allele frequency or haplotype data over sampling times to be fed as input to one-dimensional CNNs. Their method was implemented in the software , and evaluated under various sampling conditions.…”

“…Information from temporal genetic variation, either from evolve-resequence or ancient DNA (aDNA) experiments, is particularly suitable to identify when and at to what extent natural selection acted ( Dehasque et al 2020 ). Previous attempts to use deep learning to infer balancing selection from contemporary genomes ( Isildak et al 2021 ) and positive selection from temporal data ( Whitehouse and Schrider 2022 ) suggest that training an algorithm that uses the haplotype information from both contemporary and aDNA data has high potential to characterize signals of recent adaptation (and thus recent balancing selection).…”

This population does not exist: learning the distribution of evolutionary histories with generative adversarial networks

“…unknown true demographic history, low data quality, etc). Thus, this approach presents the possibility of combining the strengths of reductionist generative models with those of GANs, and could potentially be used to produce more accurate training data for machine learning-based methods for detecting selective sweeps [17,19,20,69]. Alternatively, one could take alignments from empirical data and transfer various demographic hypotheses to those alignments to help generate the reference table for performing approximate Bayesian computation.…”

“…14]. In population genetics, deep-learning methods have been used for a number of purposes, such as detecting introgression [15]-including adaptive introgression [16]; distinguishing among various modes of natural selection as well as neutral evolution [17][18][19][20]; estimating parameters such as selection coefficients and recombination rates [15,21], dispersal distances [22], and population sizes [23]; and visualizing population structure [24].…”