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

Abstract

Numerous studies over the last decade have demonstrated the utility of machine learning methods when applied to population genetic tasks. More recent studies show the potential of deep learning methods in particular, which allow researchers to approach problems without making prior assumptions about how the data should be summarized or manipulated, instead learning their own internal representation of the data in an attempt to maximize inferential accuracy. One type of deep neural network, called Generative Adversarial Networks (GANs), can even be used to generate new data, and this approach has been used to create individual artificial human genomes free from privacy concerns. In this study, we further explore the application of GANs in population genetics by designing and training a network to learn the statistical distribution of population genetic alignments (i.e. data sets consisting of sequences from an entire population sample) under several diverse evolutionary histories—the first GAN capable of performing this task. After testing multiple different neural network architectures, we report the results of a fully differentiable Deep-Convolutional Wasserstein GAN with gradient penalty that is capable of generating artificial examples of population genetic alignments that successfully mimic key aspects of the training data, including the site frequency spectrum, differentiation between populations, and patterns of linkage disequilibrium. We demonstrate consistent training success across various evolutionary models, including models of panmictic and subdivided populations, populations at equilibrium and experiencing changes in size, and populations experiencing either no selection or positive selection of various strengths, all without the need for extensive hyperparameter tuning. Overall, our findings highlight the ability of GANs to learn and mimic population genetic data and suggest future areas where this work can be applied in population genetics research that we discuss herein.

Tree sequences as a general-purpose tool for population genetic inference

“…We therefore encourage exploration of tree sequence-based deep learning inference as a general framework for population genetic inference. Our hope is that this may not only lead to improved accuracy on the problems examined here, but that tree sequence-deep learning paradigm may prove useful for a wider array of problems and, as has proved to be the case for alignment-based deep learning applications, even empower researchers to tackle new problems (Battey et al 2021;Yelmen et al 2021;Korfmann et al 2023a;Smith et al 2023;Booker et al 2023;Whitehouse and Schrider 2023).…”

“…The copyright holder for this preprint this version posted February 21, 2024. ; https://doi.org/10.1101/2024.02.20.581288 doi: bioRxiv preprint although these methods often have greater accuracy and computational efficiency (Pudlo et al 2016;Raynal et al 2019). SML methods have been applied to a variety of population genetic tasks including detecting positive selection (Pavlidis et al 2010;Lin et al 2011;Ronen et al 2013;Pybus et al 2015;Kern and Schrider 2018;Mughal and DeGiorgio 2019;Mughal et al 2020;Hejase et al 2022;Whitehouse and Schrider 2022;Arnab et al 2022), demographic parameter estimation (Sheehan and Song 2016;Wang et al 2021) and model inference (Sanchez et al 2021), inferring recombination rates (Gao et al 2016;Adrion et al 2020), detecting introgression Gower et al 2021;Ray et al 2023), and numerous other applications (Battey et al 2020(Battey et al , 2021Yelmen et al 2021;Smith et al 2023;Booker et al 2023).…”

Recent Advances in Generative Adversarial Networks for Gene Expression Data: A Comprehensive Review

“…Booker et al [68] Used GANs to learn and replicate the distribution of population genetic alignments across evolutionary histories.…”

Interpreting generative adversarial networks to infer natural selection from genetic data

“…Our previous work ( Wang et al 2021 ) described a GAN ( ) that fits an evolutionary model to any population—as training progresses, the model produces synthetic data that are closer and closer to the real data. The key innovation of is that it learns an explicit evolutionary generative model, in contrast to other GANs which generate sequences that look like real data from random processes with no underlying model ( Yelmen et al 2021 ; Booker et al 2023 ). Recent work has made use of for other species ( Small et al 2023 ) and improved the approach using adversarial Monte Carlo methods ( Gower et al 2023 ).…”