The Unreasonable Effectiveness of Convolutional Neural Networks in Population Genetic Inference

Towards the next generation of species delimitation methods: an overview of machine learning applications

Species delimitation is the process of distinguishing between populations of the same species and distinct species of a particular group of organisms. Various methods exist for inferring species limits, whether based on morphological, molecular, or other types of data. In the case of methods based on DNA sequences, most of them are rooted in the coalescent theory. However, coalescence-based models have limitations, for instance regarding complex evolutionary scenarios and large datasets. In this context, machine learning (ML) can be considered as a promising analytical tool, and provides an effective way to explore dataset structures when species-level divergences are hypothesized. In this review, we examine the use of ML in species delimitation and provide an overview and critical appraisal of existing workflows. We also provide simple explanations on how the main types of ML approaches operate, which should help uninitiated researchers and students interested in the field. Our review suggests that while current ML methods designed to infer species limits are analytically powerful, they also present specific limitations and should not be considered as definitive alternatives to coalescent methods for species delimitation. Future ML enterprises to delimit species should consider the constraints related to the use of simulated data, as in other model-based methods relying on simulations. Conversely, the flexibility of ML algorithms offers a significant advantage by enabling the analysis of diverse data types (e.g., genetic and phenotypic) and handling large datasets effectively. We also propose best practices for the use of ML methods in species delimitation, offering insights into potential future applications. We expect that the proposed guidelines will be useful for enhancing the accessibility, effectiveness, and objectivity of ML in species delimitation.

A supervised learning approach to estimating IRT models in small samples

Existing estimators of parameters of item response theory (IRT) models exploit the likelihood function. In small samples, however, the IRT likelihood oftentimes contains little informative value, potentially resulting in biased and/or unstable parameter estimates and large standard errors. To facilitate small-sample IRT estimation, we introduce a novel approach that does not rely on the likelihood. Our estimation approach derives features from response data and then maps the features to item parameters using a neural network (NN). We describe and evaluate our approach for the three-parameter logistic model; however, it is applicable to any model with an item characteristic curve. Three types of NNs are developed, supporting the obtainment of both point estimates and confidence intervals for IRT model parameters. The results of a simulation study demonstrate that these NNs perform better than Bayesian estimation using Markov chain Monte Carlo methods in terms of the quality of the point estimates and confidence intervals while also being much faster. These properties facilitate (1) pretesting items in a real-time testing environment, (2) pretesting more items and (3) pretesting items only in a secured environment to eradicate possible compromise of new items in online testing.

Efficient Detection and Characterization of Targets of Natural Selection Using Transfer Learning

Natural selection leaves detectable patterns of altered spatial diversity within genomes, and identifying affected regions is crucial for understanding species evolution. Recently, machine learning approaches applied to raw population genomic data have been developed to uncover these adaptive signatures. Convolutional neural networks (CNNs) are particularly effective for this task, as they handle large data arrays while maintaining element correlations. However, shallow CNNs may miss complex patterns due to their limited capacity, while deep CNNs can capture these patterns but require extensive data and computational power. Transfer learning addresses these challenges by utilizing a deep CNN pretrained on a large dataset as a feature extraction tool for downstream classification and evolutionary parameter prediction. This approach reduces extensive training data generation requirements and computational needs while maintaining high performance. In this study, we developed TrIdent, a tool that uses transfer learning to enhance detection of adaptive genomic regions from image representations of multilocus variation. We evaluated TrIdent across various genetic, demographic, and adaptive settings, in addition to unphased data and other confounding factors. TrIdent demonstrated improved detection of adaptive regions compared to recent methods using similar data representations. We further explored model interpretability through class activation maps and adapted TrIdent to infer selection parameters for identified adaptive candidates. Using whole-genome haplotype data from European and African populations, TrIdent effectively recapitulated known sweep candidates and identified novel cancer, and other disease-associated genes as potential sweeps.

Constructing ancestral recombination graphs through reinforcement learning

Introduction

Over the years, many approaches have been proposed to build ancestral recombination graphs (ARGs), graphs used to represent the genetic relationship between individuals. Among these methods, many rely on the assumption that the most likely graph is among those with the fewest recombination events. In this paper, we propose a new approach to build maximum parsimony ARGs: Reinforcement Learning (RL).

Methods

We exploit the similarities between finding the shortest path between a set of genetic sequences and their most recent common ancestor and finding the shortest path between the entrance and exit of a maze, a classic RL problem. In the maze problem, the learner, called the agent, must learn the directions to take in order to escape as quickly as possible, whereas in our problem, the agent must learn the actions to take between coalescence, mutation, and recombination in order to reach the most recent common ancestor as quickly as possible.

Results

Our results show that RL can be used to build ARGs with as few recombination events as those built with a heuristic algorithm optimized to build minimal ARGs, and sometimes even fewer. Moreover, our method allows to build a distribution of ARGs with few recombination events for a given sample, and can also generalize learning to new samples not used during the learning process.

Discussion

RL is a promising and innovative approach to build ARGs. By learning to construct ARGs just from the data, our method differs from conventional methods that rely on heuristic rules or complex theoretical models.

Mixed Outcomes in Recombination Rates After Domestication: Revisiting Theory and Data

The process of domestication has altered many phenotypes. Selection on these phenotypes has long been hypothesised to indirectly select for increases in the genome-wide recombination rate. This hypothesis is potentially consistent with theory on the evolution of the recombination rate, but empirical support has been unclear. We review relevant theory, lab-based experiments, and data comparing recombination rates in wild progenitors and their domesticated counterparts. We utilise population sequencing data and a deep learning method to infer genome-wide recombination rates for new comparisons of chicken/red junglefowl, sheep/mouflon, and goat/bezoar. We find evidence of increased recombination in domesticated goats compared to bezoars but more mixed results in chicken and generally decreased recombination in domesticated sheep compared to mouflon. Our results add to a growing body of literature in plants and animals that finds no consistent evidence of an increase in genome-wide recombination with domestication.

ConfuseNN: Interpreting convolutional neural network inferences in population genomics with data shuffling

Convolutional neural networks (CNNs) have become powerful tools for population genomic inference, yet understanding which genomic features drive their performance remains challenging. We introduce ConfuseNN, a method that systematically shuffles input haplotype matrices to disrupt specific population genetic features and evaluate their contribution to CNN performance. By sequentially removing signals from linkage disequilibrium, allele frequency, and other population genetic patterns in test data, we evaluate how each feature contributes to CNN performance. We applied ConfuseNN to three published CNNs for demographic history and selection inference, confirming the importance of specific data features and identifying limitations of network architecture and of simulated training and testing data design. ConfuseNN provides an accessible biologically motivated framework for interpreting CNN behavior across different tasks in population genetics, helping bridge the gap between powerful deep learning approaches and traditional population genetic theory.

Integrating Deep Learning Models with Genome-Wide Association Study-Based Identification Enhanced Phenotype Predictions in Group A Streptococcus

Group A Streptococcus (GAS) is a major pathogen with diverse clinical outcomes linked to its genetic variability, making accurate phenotype prediction essential. While previous studies have identified many GAS-associated genetic factors, translating these findings into predictive models remains challenging due to data complexity. The current study aimed to integrate deep learning models with genome-wide association study-derived genetic variants to predict pathogenic phenotypes in GAS. We evaluated the performance of several deep neural network models, including CNN, ResNet18, LSTM, and their ensemble approach in predicting GAS phenotypes. It was found that the ensemble model consistently achieved the highest prediction accuracy across phenotypes. Models trained on the full 4722-genotype set outperformed those trained on a reduced 175-genotype set, underscoring the importance of comprehensive variant data in capturing complex genotype-phenotype interactions. Performance changes in the reduced 175-genotype set compared to the full-set genotype scenarios revealed the impact of data dimensionality on model effectiveness, with CNN remaining robust, while ResNet18 and LSTM underperformed. Our findings emphasized the potential of deep learning in phenotype prediction and the critical role of data-model compatibility.

Efficient detection and characterization of targets of natural selection using transfer learning

Natural selection leaves detectable patterns of altered spatial diversity within genomes, and identifying affected regions is crucial for understanding species evolution. Recently, machine learning approaches applied to raw population genomic data have been developed to uncover these adaptive signatures. Convolutional neural networks (CNNs) are particularly effective for this task, as they handle large data arrays while maintaining element correlations. However, shallow CNNs may miss complex patterns due to their limited capacity, while deep CNNs can capture these patterns but require extensive data and computational power. Transfer learning addresses these challenges by utilizing a deep CNN pre-trained on a large dataset as a feature extraction tool for downstream classification and evolutionary parameter prediction. This approach reduces extensive training data generation requirements and computational needs while maintaining high performance. In this study, we developed TrIdent, a tool that uses transfer learning to enhance detection of adaptive genomic regions from image representations of multilocus variation. We evaluated TrIdent across various genetic, demographic, and adaptive settings, in addition to unphased data and other confounding factors. TrIdent demonstrated improved detection of adaptive regions compared to recent methods using similar data representations. We further explored model interpretability through class activation maps and adapted TrIdent to infer selection parameters for identified adaptive candidates. Using whole-genome haplotype data from European and African populations, TrIdent effectively recapitulated known sweep candidates and identified novel cancer, and other disease-associated genes as potential sweeps.

Fast and accurate deep learning scans for signatures of natural selection in genomes using FASTER-NN

Deep learning classification models based on Convolutional Neural Networks (CNNs) are increasingly used in population genetic inference for detecting signatures of natural selection. Prevailing detection methods treat the design of the classifier as a discrete phase, assuming that high classification accuracy is the sole prerequisite for precise detection. This frequently steers method development toward classification-driven optimizations that can inadvertently impede detection. We present FASTER-NN, a CNN classifier designed specifically for the precise detection of natural selection. It has higher sensitivity than state-of-the-art CNN classifiers while only processing allele frequencies and genomic positions through dilated convolutions to maximize data reuse. As a result, execution time is invariant to the sample size and the chromosome length, creating a highly suitable solution for large-scale, whole-genome scans. Furthermore, FASTER-NN can accurately identify selective sweeps in recombination hotspots, which is a highly challenging detection problem with very limited theoretical treatment to date.

Population size rescaling significantly biases outcomes of forward-in-time population genetic simulations

Simulations are an essential tool in all areas of population genetic research, used in tasks such as the validation of theoretical analysis and the study of complex evolutionary models. Forward-in-time simulations are especially flexible, allowing for various types of natural selection, complex genetic architectures, and non-Wright-Fisher dynamics. However, their intense computational requirements can be prohibitive to simulating large populations and genomes. A popular method to alleviate this burden is to scale down the population size by some scaling factor while scaling up the mutation rate, selection coefficients, and recombination rate by the same factor. However, this rescaling approach may in some cases bias simulation results. To investigate the manner and degree to which rescaling impacts simulation outcomes, we carried out simulations with different demographic histories and distributions of fitness effects using several values of the rescaling factor, Q, and compared the deviation of key outcomes (fixation times, allele frequencies, linkage disequilibrium, and the fraction of mutations that fix during the simulation) between the scaled and unscaled simulations. Our results indicate that scaling introduces substantial biases to each of these measured outcomes, even at small values of Q. Moreover, the nature of these effects depends on the evolutionary model and scaling factor being examined. While increasing the scaling factor tends to increase the observed biases, this relationship is not always straightforward; thus, it may be difficult to know the impact of scaling on simulation outcomes a priori. However, it appears that for most models, only a small number of replicates was needed to accurately quantify the bias produced by rescaling for a given Q. In summary, while rescaling forward-in-time simulations may be necessary in many cases, researchers should be aware of the rescaling procedure's impact on simulation outcomes and consider investigating its magnitude in smaller scale simulations of the desired model(s) before selecting an appropriate value of Q.

Data preprocessing methods for selective sweep detection using convolutional neural networks

The identification of positive selection has been framed as a classification task, with Convolutional Neural Networks (CNNs) already outperforming summary statistics and likelihood-based approaches in accuracy. Despite the prevalence of CNN-based methods that manipulate the pixels of images representing raw genomic data as a preprocessing step to improve classification accuracy, the efficacy of these pixel-rearrangement techniques remains inadequately examined, particularly in the presence of confounding factors like population bottlenecks, migration and recombination hotspots. We introduce a set of pixel rearrangement algorithms aimed at enhancing CNN classification accuracy in detecting selective sweeps. These algorithms are employed to assess the performance of four CNN models for selective sweep detection. Our findings illustrate that the judicious application of rearrangement algorithms notably enhances the overall classification accuracy of a CNN across various datasets simulating confounding factors. We observed that sorting the columns of the genomic matrices has higher on CNN performance than rearranging the sequences. To some extent, these rearrangement algorithms are more robust to misspecified demographic models compared with the utilization of the default preprocessing algorithm as suggested by the respective authors of each CNN architecture. We provide the data rearrangement algorithms as a distinct package available for download at: https://github.com/Zhaohq96/Genetic-data-rearrangement.

Digital Image Processing to Detect Adaptive Evolution

In recent years, advances in image processing and machine learning have fueled a paradigm shift in detecting genomic regions under natural selection. Early machine learning techniques employed population-genetic summary statistics as features, which focus on specific genomic patterns expected by adaptive and neutral processes. Though such engineered features are important when training data are limited, the ease at which simulated data can now be generated has led to the recent development of approaches that take in image representations of haplotype alignments and automatically extract important features using convolutional neural networks. Digital image processing methods termed α-molecules are a class of techniques for multiscale representation of objects that can extract a diverse set of features from images. One such α-molecule method, termed wavelet decomposition, lends greater control over high-frequency components of images. Another α-molecule method, termed curvelet decomposition, is an extension of the wavelet concept that considers events occurring along curves within images. We show that application of these α-molecule techniques to extract features from image representations of haplotype alignments yield high true positive rate and accuracy to detect hard and soft selective sweep signatures from genomic data with both linear and nonlinear machine learning classifiers. Moreover, we find that such models are easy to visualize and interpret, with performance rivaling those of contemporary deep learning approaches for detecting sweeps.

Computational Genomics and Its Applications to Anthropological Questions

The advent of affordable genome sequencing and the development of new computational tools have established a new era of genomic knowledge. Sequenced human genomes number in the tens of thousands, including thousands of ancient human genomes. The abundance of data has been met with new analysis tools that can be used to understand populations' demographic and evolutionary histories. Thus, a variety of computational methods now exist that can be leveraged to answer anthropological questions. This includes novel likelihood and Bayesian methods, machine learning techniques, and a vast array of population simulators. These computational tools provide powerful insights gained from genomic datasets, although they are generally inaccessible to those with less computational experience. Here, we outline the theoretical workings behind computational genomics methods, limitations and other considerations when applying these computational methods, and examples of how computational methods have already been applied to anthropological questions. We hope this review will empower other anthropologists to utilize these powerful tools in their own research.

Tree Sequences as a General-Purpose Tool for Population Genetic Inference

As population genetic data increase in size, new methods have been developed to store genetic information in efficient ways, such as tree sequences. These data structures are computationally and storage efficient but are not interchangeable with existing data structures used for many population genetic inference methodologies such as the use of convolutional neural networks applied to population genetic alignments. To better utilize these new data structures, we propose and implement a graph convolutional network to directly learn from tree sequence topology and node data, allowing for the use of neural network applications without an intermediate step of converting tree sequences to population genetic alignment format. We then compare our approach to standard convolutional neural network approaches on a set of previously defined benchmarking tasks including recombination rate estimation, positive selection detection, introgression detection, and demographic model parameter inference. We show that tree sequences can be directly learned from using a graph convolutional network approach and can be used to perform well on these common population genetic inference tasks with accuracies roughly matching or even exceeding that of a convolutional neural network-based method. As tree sequences become more widely used in population genetic research, we foresee developments and optimizations of this work to provide a foundation for population genetic inference moving forward.

Phylogenomics and phylogeographic model testing using convolutional neural networks reveal a history of recent admixture in the Canarian Kleinia neriifolia

Multiple-island endemics (MIE) are considered ideal natural subjects to study patterns of island colonization that involve recent population-level genetic processes. Kleinia neriifolia is a Canarian MIE widespread across the archipelago, which exhibits a close phylogenetic relationship with species in northwest Africa and at the other side of the Sahara Desert. Here, we used target sequencing with plastid skimming (Hyb-Seq), a dense population-level sampling of K. neriifolia, and representatives of its African-southern Arabian relatives to infer phylogenetic relationships and divergence times at the species and population levels. Using population genetic techniques and machine learning (convolutional neural networks [CNNs]), we reconstructed phylogeographic relationships and patterns of genetic admixture based on a multilocus SNP nuclear dataset. Phylogenomic analysis based on the nuclear dataset identifies the northwestern African Kleinia anteuphorbium as the sister species of K. neriifolia, with divergence starting in the early Pliocene. Divergence from its sister clade, comprising species from the Horn of Africa and southern Arabia, is dated to the arid Messinian period, lending support to the climatic vicariance origin of the Rand Flora. Phylogeographic model testing with CNNs supports an initial colonization of the central island of Tenerife followed by eastward and westward migration across the archipelago, which resulted in the observed east/west phylogeographic split. Subsequent population extinctions linked to aridification events, and recolonization from Tenerife, are proposed to explain the patterns of genetic admixture in the eastern Canary Islands. We demonstrate that CNNs based on SNPs can be used to discriminate among complex scenarios of island migration and colonization.

A Next Generation of Hierarchical Bayesian Analyses of Hybrid Zones Enables Model-Based Quantification of Variation in Introgression in R

Hybrid zones, where genetically distinct groups of organisms meet and interbreed, offer valuable insights into the nature of species and speciation. Here, we present a new R package, bgchm, for population genomic analyses of hybrid zones. This R package extends and updates the existing bgc software and combines Bayesian analyses of hierarchical genomic clines with Bayesian methods for estimating hybrid indexes, interpopulation ancestry proportions, and geographic clines. Compared to existing software, bgchm offers enhanced efficiency through Hamiltonian Monte Carlo sampling and the ability to work with genotype likelihoods combined with a hierarchical Bayesian approach, enabling inference for diverse types of genetic data sets. The package also facilitates the quantification of introgression patterns across genomes, which is crucial for understanding reproductive isolation and speciation genetics. We first describe the models underlying bgchm and then provide an overview of the R package and illustrate its use through the analysis of simulated and empirical data sets. We show that bgchm generates accurate estimates of model parameters under a variety of conditions, especially when the genetic loci analyzed are highly ancestry informative. This includes relatively robust estimates of genome-wide variability in clines, which has not been the focus of previous models and methods. We also illustrate how both selection and genetic drift contribute to variability in introgression among loci and how additional information can be used to help distinguish these contributions. We conclude by describing the promises and limitations of bgchm, comparing bgchm to other software for genomic cline analyses, and identifying areas for fruitful future development.

Population Genomic Scans for Natural Selection and Demography

Uncovering the fundamental processes that shape genomic variation in natural populations is a primary objective of population genetics. These processes include demographic effects such as past changes in effective population size or gene flow between structured populations. Furthermore, genomic variation is affected by selection on nonneutral genetic variants, for example, through the adaptation of beneficial alleles or balancing selection that maintains genetic variation. In this article, we discuss the characterization of these processes using population genetic models, and we review methods developed on the basis of these models to unravel the underlying processes from modern population genomic data sets. We briefly discuss the conditions in which these approaches can be used to infer demography or identify specific nonneutral genetic variants and cases in which caution is warranted. Moreover, we summarize the challenges of jointly inferring demography and selective processes that affect neutral variation genome-wide.

Deep estimation of the intensity and timing of natural selection from ancient genomes

Leveraging past allele frequencies has proven to be key for identifying the impact of natural selection across time. However, this approach suffers from imprecise estimations of the intensity (s) and timing (T) of selection, particularly when ancient samples are scarce in specific epochs. Here, we aimed to bypass the computation of allele frequencies across arbitrarily defined past epochs and refine the estimations of selection parameters by implementing convolutional neural networks (CNNs) algorithms that directly use ancient genotypes sampled across time. Using computer simulations, we first show that genotype-based CNNs consistently outperform an approximate Bayesian computation (ABC) approach based on past allele frequency trajectories, regardless of the selection model assumed and the number of available ancient genotypes. When applying this method to empirical data from modern and ancient Europeans, we replicated the reported increased number of selection events in post-Neolithic Europe, independently of the continental subregion studied. Furthermore, we substantially refined the ABC-based estimations of s and T for a set of positively and negatively selected variants, including iconic cases of positive selection and experimentally validated disease-risk variants. Our CNN predictions support a history of recent positive and negative selection targeting variants associated with host defence against pathogens, aligning with previous work that highlights the significant impact of infectious diseases, such as tuberculosis, in Europe. These findings collectively demonstrate that detecting the footprints of natural selection on ancient genomes is crucial for unravelling the history of severe human diseases.

Toward a Semi-Supervised Learning Approach to Phylogenetic Estimation

Models have always been central to inferring molecular evolution and to reconstructing phylogenetic trees. Their use typically involves the development of a mechanistic framework reflecting our understanding of the underlying biological processes, such as nucleotide substitutions, and the estimation of model parameters by maximum likelihood or Bayesian inference. However, deriving and optimizing the likelihood of the data is not always possible under complex evolutionary scenarios or even tractable for large datasets, often leading to unrealistic simplifying assumptions in the fitted models. To overcome this issue, we coupled stochastic simulations of genome evolution with a new supervised deep-learning model to infer key parameters of molecular evolution. Our model is designed to directly analyze multiple sequence alignments and estimate per-site evolutionary rates and divergence without requiring a known phylogenetic tree. The accuracy of our predictions matched that of likelihood-based phylogenetic inference when rate heterogeneity followed a simple gamma distribution, but it strongly exceeded it under more complex patterns of rate variation, such as codon models. Our approach is highly scalable and can be efficiently applied to genomic data, as we showed on a dataset of 26 million nucleotides from the clownfish clade. Our simulations also showed that the integration of per-site rates obtained by deep learning within a Bayesian framework led to significantly more accurate phylogenetic inference, particularly with respect to the estimated branch lengths. We thus propose that future advancements in phylogenetic analysis will benefit from a semi-supervised learning approach that combines deep-learning estimation of substitution rates, which allows for more flexible models of rate variation, and probabilistic inference of the phylogenetic tree, which guarantees interpretability and a rigorous assessment of statistical support.

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

As population genetics data increases in size new methods have been developed to store genetic information in efficient ways, such as tree sequences. These data structures are computationally and storage efficient, but are not interchangeable with existing data structures used for many population genetic inference methodologies such as the use of convolutional neural networks (CNNs) applied to population genetic alignments. To better utilize these new data structures we propose and implement a graph convolutional network (GCN) to directly learn from tree sequence topology and node data, allowing for the use of neural network applications without an intermediate step of converting tree sequences to population genetic alignment format. We then compare our approach to standard CNN approaches on a set of previously defined benchmarking tasks including recombination rate estimation, positive selection detection, introgression detection, and demographic model parameter inference. We show that tree sequences can be directly learned from using a GCN approach and can be used to perform well on these common population genetics inference tasks with accuracies roughly matching or even exceeding that of a CNN-based method. As tree sequences become more widely used in population genetics research we foresee developments and optimizations of this work to provide a foundation for population genetics inference moving forward.

Population size rescaling significantly biases outcomes of forward-in-time population genetic simulations

Simulations are an essential tool in all areas of population genetic research, used in tasks such as the validation of theoretical analysis and the study of complex evolutionary models. Forward-in-time simulations are especially flexible, allowing for various types of natural selection, complex genetic architectures, and non-Wright-Fisher dynamics. However, their intense computational requirements can be prohibitive to simulating large populations and genomes. A popular method to alleviate this burden is to scale down the population size by some scaling factor while scaling up the mutation rate, selection coefficients, and recombination rate by the same factor. However, this rescaling approach may in some cases bias simulation results. To investigate the manner and degree to which rescaling impacts simulation outcomes, we carried out simulations with different demographic histories and distributions of fitness effects using several values of the rescaling factor, Q , and compared the deviation of key outcomes (fixation times, allele frequencies, linkage disequilibrium, and the fraction of mutations that fix during the simulation) between the scaled and unscaled simulations. Our results indicate that scaling introduces substantial biases to each of these measured outcomes, even at small values of Q . Moreover, the nature of these effects depends on the evolutionary model and scaling factor being examined. While increasing the scaling factor tends to increase the observed biases, this relationship is not always straightforward, thus it may be difficult to know the impact of scaling on simulation outcomes a priori. However, it appears that for most models, only a small number of replicates was needed to accurately quantify the bias produced by rescaling for a given Q . In summary, while rescaling forward-in-time simulations may be necessary in many cases, researchers should be aware of the rescaling procedure's impact on simulation outcomes and consider investigating its magnitude in smaller scale simulations of the desired model(s) before selecting an appropriate value of Q .

Scalable CNN-based classification of selective sweeps using derived allele frequencies

MOTIVATION

Selective sweeps can successfully be distinguished from neutral genetic data using summary statistics and likelihood-based methods that analyze single nucleotide polymorphisms (SNPs). However, these methods are sensitive to confounding factors, such as severe population bottlenecks and old migration. By virtue of machine learning, and specifically convolutional neural networks (CNNs), new accurate classification models that are robust to confounding factors have been recently proposed. However, such methods are more computationally expensive than summary-statistic-based ones, yielding them impractical for processing large-scale genomic data. Moreover, SNP data are frequently preprocessed to improve classification accuracy, further exacerbating the long analysis times.

RESULTS

To this end, we propose a 1D CNN-based model, dubbed FAST-NN, that does not require any preprocessing while using only derived allele frequencies instead of summary statistics or raw SNP data, thereby yielding a sample-size-invariant, scalable solution. We evaluated several data fusion approaches to account for the variance of the density of genetic diversity across genomic regions (a selective sweep signature), and performed an extensive neural architecture search based on a state-of-the-art reference network architecture (SweepNet). The resulting model, FAST-NN, outperforms the reference architecture by up to 12% inference accuracy over all challenging evolutionary scenarios with confounding factors that were evaluated. Moreover, FAST-NN is between 30× and 259× faster on a single CPU core, and between 2.0× and 6.2× faster on a GPU, when processing sample sizes between 128 and 1000 samples. Our work paves the way for the practical use of CNNs in large-scale selective sweep detection.

AVAILABILITY AND IMPLEMENTATION

https://github.com/SjoerdvandenBelt/FAST-NN.

Artificial intelligence enables unified analysis of historical and landscape influences on genetic diversity

While genetic variation in any species is potentially shaped by a range of processes, phylogeography and landscape genetics are largely concerned with inferring how environmental conditions and landscape features impact neutral intraspecific diversity. However, even as both disciplines have come to utilize SNP data over the last decades, analytical approaches have remained for the most part focused on either broad-scale inferences of historical processes (phylogeography) or on more localized inferences about environmental and/or landscape features (landscape genetics). Here we demonstrate that an artificial intelligence model-based analytical framework can consider both deeper historical factors and landscape-level processes in an integrated analysis. We implement this framework using data collected from two Brazilian anurans, the Brazilian sibilator frog (Leptodactylus troglodytes) and granular toad (Rhinella granulosa). Our results indicate that historical demographic processes shape most the genetic variation in the sibulator frog, while landscape processes primarily influence variation in the granular toad. The machine learning framework used here allows both historical and landscape processes to be considered equally, rather than requiring researchers to make an a priori decision about which factors are important.

Applications of machine learning in phylogenetics

Machine learning has increasingly been applied to a wide range of questions in phylogenetic inference. Supervised machine learning approaches that rely on simulated training data have been used to infer tree topologies and branch lengths, to select substitution models, and to perform downstream inferences of introgression and diversification. Here, we review how researchers have used several promising machine learning approaches to make phylogenetic inferences. Despite the promise of these methods, several barriers prevent supervised machine learning from reaching its full potential in phylogenetics. We discuss these barriers and potential paths forward. In the future, we expect that the application of careful network designs and data encodings will allow supervised machine learning to accommodate the complex processes that continue to confound traditional phylogenetic methods.

Genetic structure and landscape effects on gene flow in the Neotropical lizard Norops brasiliensis (Squamata: Dactyloidae)

One key research goal of evolutionary biology is to understand the origin and maintenance of genetic variation. In the Cerrado, the South American savanna located primarily in the Central Brazilian Plateau, many hypotheses have been proposed to explain how landscape features (e.g., geographic distance, river barriers, topographic compartmentalization, and historical climatic fluctuations) have promoted genetic structure by mediating gene flow. Here, we asked whether these landscape features have influenced the genetic structure and differentiation in the lizard species Norops brasiliensis (Squamata: Dactyloidae). To achieve our goal, we used a genetic clustering analysis and estimate an effective migration surface to assess genetic structure in the focal species. Optimized isolation-by-resistance models and a simulation-based approach combined with machine learning (convolutional neural network; CNN) were then used to infer current and historical effects on population genetic structure through 12 unique landscape models. We recovered five geographically distributed populations that are separated by regions of lower-than-expected gene flow. The results of the CNN showed that geographic distance is the sole predictor of genetic variation in N. brasiliensis, and that slope, rivers, and historical climate had no discernible influence on gene flow. Our novel CNN approach was accurate (89.5%) in differentiating each landscape model. CNN and other machine learning approaches are still largely unexplored in landscape genetics studies, representing promising avenues for future research with increasingly accessible genomic datasets.

Deep Learning and Likelihood Approaches for Viral Phylogeography Converge on the Same Answers Whether the Inference Model Is Right or Wrong

Analysis of phylogenetic trees has become an essential tool in epidemiology. Likelihood-based methods fit models to phylogenies to draw inferences about the phylodynamics and history of viral transmission. However, these methods are often computationally expensive, which limits the complexity and realism of phylodynamic models and makes them ill-suited for informing policy decisions in real-time during rapidly developing outbreaks. Likelihood-free methods using deep learning are pushing the boundaries of inference beyond these constraints. In this paper, we extend, compare, and contrast a recently developed deep learning method for likelihood-free inference from trees. We trained multiple deep neural networks using phylogenies from simulated outbreaks that spread among 5 locations and found they achieve close to the same levels of accuracy as Bayesian inference under the true simulation model. We compared robustness to model misspecification of a trained neural network to that of a Bayesian method. We found that both models had comparable performance, converging on similar biases. We also implemented a method of uncertainty quantification called conformalized quantile regression that we demonstrate has similar patterns of sensitivity to model misspecification as Bayesian highest posterior density (HPD) and greatly overlap with HPDs, but have lower precision (more conservative). Finally, we trained and tested a neural network against phylogeographic data from a recent study of the SARS-Cov-2 pandemic in Europe and obtained similar estimates of region-specific epidemiological parameters and the location of the common ancestor in Europe. Along with being as accurate and robust as likelihood-based methods, our trained neural networks are on average over 3 orders of magnitude faster after training. Our results support the notion that neural networks can be trained with simulated data to accurately mimic the good and bad statistical properties of the likelihood functions of generative phylogenetic models.

The sweet tabaiba or there and back again: phylogeographical history of the Macaronesian Euphorbia balsamifera

BACKGROUND AND AIMS

Biogeographical relationships between the Canary Islands and north-west Africa are often explained by oceanic dispersal and geographical proximity. Sister-group relationships between Canarian and eastern African/Arabian taxa, the 'Rand Flora' pattern, are rare among plants and have been attributed to the extinction of north-western African populations. Euphorbia balsamifera is the only representative species of this pattern that is distributed in the Canary Islands and north-west Africa; it is also one of few species present in all seven islands. Previous studies placed African populations of E. balsamifera as sister to the Canarian populations, but this relationship was based on herbarium samples with highly degraded DNA. Here, we test the extinction hypothesis by sampling new continental populations; we also expand the Canarian sampling to examine the dynamics of island colonization and diversification.

METHODS

Using target enrichment with genome skimming, we reconstructed phylogenetic relationships within E. balsamifera and between this species and its disjunct relatives. A single nucleotide polymorphism dataset obtained from the target sequences was used to infer population genetic diversity patterns. We used convolutional neural networks to discriminate among alternative Canary Islands colonization scenarios.

KEY RESULTS

The results confirmed the Rand Flora sister-group relationship between western E. balsamifera and Euphorbia adenensis in the Eritreo-Arabian region and recovered an eastern-western geographical structure among E. balsamifera Canarian populations. Convolutional neural networks supported a scenario of east-to-west island colonization, followed by population extinctions in Lanzarote and Fuerteventura and recolonization from Tenerife and Gran Canaria; a signal of admixture between the eastern island and north-west African populations was recovered.

CONCLUSIONS

Our findings support the Surfing Syngameon Hypothesis for the colonization of the Canary Islands by E. balsamifera, but also a recent back-colonization to the continent. Populations of E. balsamifera from northwest Africa are not the remnants of an ancestral continental stock, but originated from migration events from Lanzarote and Fuerteventura. This is further evidence that oceanic archipelagos are not a sink for biodiversity, but may be a source of new genetic variability.

Computationally Efficient Demographic History Inference from Allele Frequencies with Supervised Machine Learning

Inferring past demographic history of natural populations from genomic data is of central concern in many studies across research fields. Previously, our group had developed dadi, a widely used demographic history inference method based on the allele frequency spectrum (AFS) and maximum composite-likelihood optimization. However, dadi's optimization procedure can be computationally expensive. Here, we present donni (demography optimization via neural network inference), a new inference method based on dadi that is more efficient while maintaining comparable inference accuracy. For each dadi-supported demographic model, donni simulates the expected AFS for a range of model parameters then trains a set of Mean Variance Estimation neural networks using the simulated AFS. Trained networks can then be used to instantaneously infer the model parameters from future genomic data summarized by an AFS. We demonstrate that for many demographic models, donni can infer some parameters, such as population size changes, very well and other parameters, such as migration rates and times of demographic events, fairly well. Importantly, donni provides both parameter and confidence interval estimates from input AFS with accuracy comparable to parameters inferred by dadi's likelihood optimization while bypassing its long and computationally intensive evaluation process. donni's performance demonstrates that supervised machine learning algorithms may be a promising avenue for developing more sustainable and computationally efficient demographic history inference methods.

Genotype sampling for deep-learning assisted experimental mapping of a combinatorially complete fitness landscape

MOTIVATION

Experimental characterization of fitness landscapes, which map genotypes onto fitness, is important for both evolutionary biology and protein engineering. It faces a fundamental obstacle in the astronomical number of genotypes whose fitness needs to be measured for any one protein. Deep learning may help to predict the fitness of many genotypes from a smaller neural network training sample of genotypes with experimentally measured fitness. Here I use a recently published experimentally mapped fitness landscape of more than 260 000 protein genotypes to ask how such sampling is best performed.

RESULTS

I show that multilayer perceptrons, recurrent neural networks, convolutional networks, and transformers, can explain more than 90% of fitness variance in the data. In addition, 90% of this performance is reached with a training sample comprising merely ≈103 sequences. Generalization to unseen test data is best when training data is sampled randomly and uniformly, or sampled to minimize the number of synonymous sequences. In contrast, sampling to maximize sequence diversity or codon usage bias reduces performance substantially. These observations hold for more than one network architecture. Simple sampling strategies may perform best when training deep learning neural networks to map fitness landscapes from experimental data.

AVAILABILITY AND IMPLEMENTATION

The fitness landscape data analyzed here is publicly available as described previously (Papkou et al. 2023). All code used to analyze this landscape is publicly available at https://github.com/andreas-wagner-uzh/fitness_landscape_sampling.

Interpreting generative adversarial networks to infer natural selection from genetic data

Understanding natural selection and other forms of non-neutrality is a major focus for the use of machine learning in population genetics. Existing methods rely on computationally intensive simulated training data. Unlike efficient neutral coalescent simulations for demographic inference, realistic simulations of selection typically require slow forward simulations. Because there are many possible modes of selection, a high dimensional parameter space must be explored, with no guarantee that the simulated models are close to the real processes. Finally, it is difficult to interpret trained neural networks, leading to a lack of understanding about what features contribute to classification. Here we develop a new approach to detect selection and other local evolutionary processes that requires relatively few selection simulations during training. We build upon a generative adversarial network trained to simulate realistic neutral data. This consists of a generator (fitted demographic model), and a discriminator (convolutional neural network) that predicts whether a genomic region is real or fake. As the generator can only generate data under neutral demographic processes, regions of real data that the discriminator recognizes as having a high probability of being "real" do not fit the neutral demographic model and are therefore candidates for targets of selection. To incentivize identification of a specific mode of selection, we fine-tune the discriminator with a small number of custom non-neutral simulations. We show that this approach has high power to detect various forms of selection in simulations, and that it finds regions under positive selection identified by state-of-the-art population genetic methods in three human populations. Finally, we show how to interpret the trained networks by clustering hidden units of the discriminator based on their correlation patterns with known summary statistics.

A Novel Approach Utilizing Domain Adversarial Neural Networks for the Detection and Classification of Selective Sweeps

The identification and classification of selective sweeps are of great significance for improving the understanding of biological evolution and exploring opportunities for precision medicine and genetic improvement. Here, a domain adaptation sweep detection and classification (DASDC) method is presented to balance the alignment of two domains and the classification performance through a domain-adversarial neural network and its adversarial learning modules. DASDC effectively addresses the issue of mismatch between training data and real genomic data in deep learning models, leading to a significant improvement in its generalization capability, prediction robustness, and accuracy. The DASDC method demonstrates improved identification performance compared to existing methods and excels in classification performance, particularly in scenarios where there is a mismatch between application data and training data. The successful implementation of DASDC in real data of three distinct species highlights its potential as a useful tool for identifying crucial functional genes and investigating adaptive evolutionary mechanisms, particularly with the increasing availability of genomic data.

Computationally efficient demographic history inference from allele frequencies with supervised machine learning

Inferring past demographic history of natural populations from genomic data is of central concern in many studies across research fields. Previously, our group had developed dadi, a widely used demographic history inference method based on the allele frequency spectrum (AFS) and maximum composite likelihood optimization. However, dadi's optimization procedure can be computationally expensive. Here, we developed donni (demography optimization via neural network inference), a new inference method based on dadi that is more efficient while maintaining comparable inference accuracy. For each dadi-supported demographic model, donni simulates the expected AFS for a range of model parameters then trains a set of Mean Variance Estimation neural networks using the simulated AFS. Trained networks can then be used to instantaneously infer the model parameters from future input data AFS. We demonstrated that for many demographic models, donni can infer some parameters, such as population size changes, very well and other parameters, such as migration rates and times of demographic events, fairly well. Importantly, donni provides both parameter and confidence interval estimates from input AFS with accuracy comparable to parameters inferred by dadi's likelihood optimization while bypassing its long and computationally intensive evaluation process. donni's performance demonstrates that supervised machine learning algorithms may be a promising avenue for developing more sustainable and computationally efficient demographic history inference methods.

IntroUNET: Identifying introgressed alleles via semantic segmentation

A growing body of evidence suggests that gene flow between closely related species is a widespread phenomenon. Alleles that introgress from one species into a close relative are typically neutral or deleterious, but sometimes confer a significant fitness advantage. Given the potential relevance to speciation and adaptation, numerous methods have therefore been devised to identify regions of the genome that have experienced introgression. Recently, supervised machine learning approaches have been shown to be highly effective for detecting introgression. One especially promising approach is to treat population genetic inference as an image classification problem, and feed an image representation of a population genetic alignment as input to a deep neural network that distinguishes among evolutionary models (i.e. introgression or no introgression). However, if we wish to investigate the full extent and fitness effects of introgression, merely identifying genomic regions in a population genetic alignment that harbor introgressed loci is insufficient-ideally we would be able to infer precisely which individuals have introgressed material and at which positions in the genome. Here we adapt a deep learning algorithm for semantic segmentation, the task of correctly identifying the type of object to which each individual pixel in an image belongs, to the task of identifying introgressed alleles. Our trained neural network is thus able to infer, for each individual in a two-population alignment, which of those individual's alleles were introgressed from the other population. We use simulated data to show that this approach is highly accurate, and that it can be readily extended to identify alleles that are introgressed from an unsampled "ghost" population, performing comparably to a supervised learning method tailored specifically to that task. Finally, we apply this method to data from Drosophila, showing that it is able to accurately recover introgressed haplotypes from real data. This analysis reveals that introgressed alleles are typically confined to lower frequencies within genic regions, suggestive of purifying selection, but are found at much higher frequencies in a region previously shown to be affected by adaptive introgression. Our method's success in recovering introgressed haplotypes in challenging real-world scenarios underscores the utility of deep learning approaches for making richer evolutionary inferences from genomic data.

IntroUNET: identifying introgressed alleles via semantic segmentation

A growing body of evidence suggests that gene flow between closely related species is a widespread phenomenon. Alleles that introgress from one species into a close relative are typically neutral or deleterious, but sometimes confer a significant fitness advantage. Given the potential relevance to speciation and adaptation, numerous methods have therefore been devised to identify regions of the genome that have experienced introgression. Recently, supervised machine learning approaches have been shown to be highly effective for detecting introgression. One especially promising approach is to treat population genetic inference as an image classification problem, and feed an image representation of a population genetic alignment as input to a deep neural network that distinguishes among evolutionary models (i.e. introgression or no introgression). However, if we wish to investigate the full extent and fitness effects of introgression, merely identifying genomic regions in a population genetic alignment that harbor introgressed loci is insufficient-ideally we would be able to infer precisely which individuals have introgressed material and at which positions in the genome. Here we adapt a deep learning algorithm for semantic segmentation, the task of correctly identifying the type of object to which each individual pixel in an image belongs, to the task of identifying introgressed alleles. Our trained neural network is thus able to infer, for each individual in a two-population alignment, which of those individual's alleles were introgressed from the other population. We use simulated data to show that this approach is highly accurate, and that it can be readily extended to identify alleles that are introgressed from an unsampled "ghost" population, performing comparably to a supervised learning method tailored specifically to that task. Finally, we apply this method to data from Drosophila, showing that it is able to accurately recover introgressed haplotypes from real data. This analysis reveals that introgressed alleles are typically confined to lower frequencies within genic regions, suggestive of purifying selection, but are found at much higher frequencies in a region previously shown to be affected by adaptive introgression. Our method's success in recovering introgressed haplotypes in challenging real-world scenarios underscores the utility of deep learning approaches for making richer evolutionary inferences from genomic data.

Forty Years of Inferential Methods in the Journals of the Society for Molecular Biology and Evolution

We are launching a series to celebrate the 40th anniversary of the first issue of Molecular Biology and Evolution. In 2024, we will publish virtual issues containing selected papers published in the Society for Molecular Biology and Evolution journals, Molecular Biology and Evolution and Genome Biology and Evolution. Each virtual issue will be accompanied by a perspective that highlights the historic and contemporary contributions of our journals to a specific topic in molecular evolution. This perspective, the first in the series, presents an account of the broad array of methods that have been published in the Society for Molecular Biology and Evolution journals, including methods to infer phylogenies, to test hypotheses in a phylogenetic framework, and to infer population genetic processes. We also mention many of the software implementations that make methods tractable for empiricists. In short, the Society for Molecular Biology and Evolution community has much to celebrate after four decades of publishing high-quality science including numerous important inferential methods.

Harnessing deep learning for population genetic inference

In population genetics, the emergence of large-scale genomic data for various species and populations has provided new opportunities to understand the evolutionary forces that drive genetic diversity using statistical inference. However, the era of population genomics presents new challenges in analysing the massive amounts of genomes and variants. Deep learning has demonstrated state-of-the-art performance for numerous applications involving large-scale data. Recently, deep learning approaches have gained popularity in population genetics; facilitated by the advent of massive genomic data sets, powerful computational hardware and complex deep learning architectures, they have been used to identify population structure, infer demographic history and investigate natural selection. Here, we introduce common deep learning architectures and provide comprehensive guidelines for implementing deep learning models for population genetic inference. We also discuss current challenges and future directions for applying deep learning in population genetics, focusing on efficiency, robustness and interpretability.

Deep Learning from Phylogenies for Diversification Analyses

Birth-death (BD) models are widely used in combination with species phylogenies to study past diversification dynamics. Current inference approaches typically rely on likelihood-based methods. These methods are not generalizable, as a new likelihood formula must be established each time a new model is proposed; for some models, such a formula is not even tractable. Deep learning can bring solutions in such situations, as deep neural networks can be trained to learn the relation between simulations and parameter values as a regression problem. In this paper, we adapt a recently developed deep learning method from pathogen phylodynamics to the case of diversification inference, and we extend its applicability to the case of the inference of state-dependent diversification models from phylogenies associated with trait data. We demonstrate the accuracy and time efficiency of the approach for the time-constant homogeneous BD model and the Binary-State Speciation and Extinction model. Finally, we illustrate the use of the proposed inference machinery by reanalyzing a phylogeny of primates and their associated ecological role as seed dispersers. Deep learning inference provides at least the same accuracy as likelihood-based inference while being faster by several orders of magnitude, offering a promising new inference approach for the deployment of future models in the field.

The Impact of Chromosomal Rearrangements in Speciation: From Micro- to Macroevolution

Chromosomal rearrangements (CRs) have been known since almost the beginning of genetics. While an important role for CRs in speciation has been suggested, evidence primarily stems from theoretical and empirical studies focusing on the microevolutionary level (i.e., on taxon pairs where speciation is often incomplete). Although the role of CRs in eukaryotic speciation at a macroevolutionary level has been supported by associations between species diversity and rates of evolution of CRs across phylogenies, these findings are limited to a restricted range of CRs and taxa. Now that more broadly applicable and precise CR detection approaches have become available, we address the challenges in filling some of the conceptual and empirical gaps between micro- and macroevolutionary studies on the role of CRs in speciation. We synthesize what is known about the macroevolutionary impact of CRs and suggest new research avenues to overcome the pitfalls of previous studies to gain a more comprehensive understanding of the evolutionary significance of CRs in speciation across the tree of life.

Inferring Historical Introgression with Deep Learning

Resolving phylogenetic relationships among taxa remains a challenge in the era of big data due to the presence of genetic admixture in a wide range of organisms. Rapidly developing sequencing technologies and statistical tests enable evolutionary relationships to be disentangled at a genome-wide level, yet many of these tests are computationally intensive and rely on phased genotypes, large sample sizes, restricted phylogenetic topologies, or hypothesis testing. To overcome these difficulties, we developed a deep learning-based approach, named ERICA, for inferring genome-wide evolutionary relationships and local introgressed regions from sequence data. ERICA accepts sequence alignments of both population genomic data and multiple genome assemblies, and efficiently identifies discordant genealogy patterns and exchanged regions across genomes when compared with other methods. We further tested ERICA using real population genomic data from Heliconius butterflies that have undergone adaptive radiation and frequent hybridization. Finally, we applied ERICA to characterize hybridization and introgression in wild and cultivated rice, revealing the important role of introgression in rice domestication and adaptation. Taken together, our findings demonstrate that ERICA provides an effective method for teasing apart evolutionary relationships using whole genome data, which can ultimately facilitate evolutionary studies on hybridization and introgression.

On convolutional neural networks for selection inference: Revealing the effect of preprocessing on model learning and the capacity to discover novel patterns

A central challenge in population genetics is the detection of genomic footprints of selection. As machine learning tools including convolutional neural networks (CNNs) have become more sophisticated and applied more broadly, these provide a logical next step for increasing our power to learn and detect such patterns; indeed, CNNs trained on simulated genome sequences have recently been shown to be highly effective at this task. Unlike previous approaches, which rely upon human-crafted summary statistics, these methods are able to be applied directly to raw genomic data, allowing them to potentially learn new signatures that, if well-understood, could improve the current theory surrounding selective sweeps. Towards this end, we examine a representative CNN from the literature, paring it down to the minimal complexity needed to maintain comparable performance; this low-complexity CNN allows us to directly interpret the learned evolutionary signatures. We then validate these patterns in more complex models using metrics that evaluate feature importance. Our findings reveal that preprocessing steps, which determine how the population genetic data is presented to the model, play a central role in the learned prediction method. This results in models that mimic previously-defined summary statistics; in one case, the summary statistic itself achieves similarly high accuracy. For evolutionary processes that are less well understood than selective sweeps, we hope this provides an initial framework for using CNNs in ways that go beyond simply achieving high classification performance. Instead, we propose that CNNs might be useful as tools for learning novel patterns that can translate to easy-to-implement summary statistics available to a wider community of researchers.

Domain-adaptive neural networks improve supervised machine learning based on simulated population genetic data

Investigators have recently introduced powerful methods for population genetic inference that rely on supervised machine learning from simulated data. Despite their performance advantages, these methods can fail when the simulated training data does not adequately resemble data from the real world. Here, we show that this "simulation mis-specification" problem can be framed as a "domain adaptation" problem, where a model learned from one data distribution is applied to a dataset drawn from a different distribution. By applying an established domain-adaptation technique based on a gradient reversal layer (GRL), originally introduced for image classification, we show that the effects of simulation mis-specification can be substantially mitigated. We focus our analysis on two state-of-the-art deep-learning population genetic methods-SIA, which infers positive selection from features of the ancestral recombination graph (ARG), and ReLERNN, which infers recombination rates from genotype matrices. In the case of SIA, the domain adaptive framework also compensates for ARG inference error. Using the domain-adaptive SIA (dadaSIA) model, we estimate improved selection coefficients at selected loci in the 1000 Genomes CEU population. We anticipate that domain adaptation will prove to be widely applicable in the growing use of supervised machine learning in population genetics.

Tensor Decomposition-based Feature Extraction and Classification to Detect Natural Selection from Genomic Data

Inferences of adaptive events are important for learning about traits, such as human digestion of lactose after infancy and the rapid spread of viral variants. Early efforts toward identifying footprints of natural selection from genomic data involved development of summary statistic and likelihood methods. However, such techniques are grounded in simple patterns or theoretical models that limit the complexity of settings they can explore. Due to the renaissance in artificial intelligence, machine learning methods have taken center stage in recent efforts to detect natural selection, with strategies such as convolutional neural networks applied to images of haplotypes. Yet, limitations of such techniques include estimation of large numbers of model parameters under nonconvex settings and feature identification without regard to location within an image. An alternative approach is to use tensor decomposition to extract features from multidimensional data although preserving the latent structure of the data, and to feed these features to machine learning models. Here, we adopt this framework and present a novel approach termed T-REx, which extracts features from images of haplotypes across sampled individuals using tensor decomposition, and then makes predictions from these features using classical machine learning methods. As a proof of concept, we explore the performance of T-REx on simulated neutral and selective sweep scenarios and find that it has high power and accuracy to discriminate sweeps from neutrality, robustness to common technical hurdles, and easy visualization of feature importance. Therefore, T-REx is a powerful addition to the toolkit for detecting adaptive processes from genomic data.

Domain-adaptive neural networks improve supervised machine learning based on simulated population genetic data

Investigators have recently introduced powerful methods for population genetic inference that rely on supervised machine learning from simulated data. Despite their performance advantages, these methods can fail when the simulated training data does not adequately resemble data from the real world. Here, we show that this "simulation mis-specification" problem can be framed as a "domain adaptation" problem, where a model learned from one data distribution is applied to a dataset drawn from a different distribution. By applying an established domain-adaptation technique based on a gradient reversal layer (GRL), originally introduced for image classification, we show that the effects of simulation mis-specification can be substantially mitigated. We focus our analysis on two state-of-the-art deep-learning population genetic methods-SIA, which infers positive selection from features of the ancestral recombination graph (ARG), and ReLERNN, which infers recombination rates from genotype matrices. In the case of SIA, the domain adaptive framework also compensates for ARG inference error. Using the domain-adaptive SIA (dadaSIA) model, we estimate improved selection coefficients at selected loci in the 1000 Genomes CEU population. We anticipate that domain adaptation will prove to be widely applicable in the growing use of supervised machine learning in population genetics.

A few genetic variants go a long way in differentiating Penstemon species

The integrity of hybridizing species is usually maintained by genome-wide selection or by selection on a few genomic regions. A study published in PLOS Biology finds a different pattern-60 SNPs spread across the genome differentiate a Penstemon species pair.

Genes selection using deep learning and explainable artificial intelligence for chronic lymphocytic leukemia predicting the need and time to therapy

Analyzing gene expression profiles (GEP) through artificial intelligence provides meaningful insight into cancer disease. This study introduces DeepSHAP Autoencoder Filter for Genes Selection (DSAF-GS), a novel deep learning and explainable artificial intelligence-based approach for feature selection in genomics-scale data. DSAF-GS exploits the autoencoder's reconstruction capabilities without changing the original feature space, enhancing the interpretation of the results. Explainable artificial intelligence is then used to select the informative genes for chronic lymphocytic leukemia prognosis of 217 cases from a GEP database comprising roughly 20,000 genes. The model for prognosis prediction achieved an accuracy of 86.4%, a sensitivity of 85.0%, and a specificity of 87.5%. According to the proposed approach, predictions were strongly influenced by CEACAM19 and PIGP, moderately influenced by MKL1 and GNE, and poorly influenced by other genes. The 10 most influential genes were selected for further analysis. Among them, FADD, FIBP, FIBP, GNE, IGF1R, MKL1, PIGP, and SLC39A6 were identified in the Reactome pathway database as involved in signal transduction, transcription, protein metabolism, immune system, cell cycle, and apoptosis. Moreover, according to the network model of the 3D protein-protein interaction (PPI) explored using the NetworkAnalyst tool, FADD, FIBP, IGF1R, QTRT1, GNE, SLC39A6, and MKL1 appear coupled into a complex network. Finally, all 10 selected genes showed a predictive power on time to first treatment (TTFT) in univariate analyses on a basic prognostic model including IGHV mutational status, del(11q) and del(17p), NOTCH1 mutations, β2-microglobulin, Rai stage, and B-lymphocytosis known to predict TTFT in CLL. However, only IGF1R [hazard ratio (HR) 1.41, 95% CI 1.08-1.84, P=0.013), COL28A1 (HR 0.32, 95% CI 0.10-0.97, P=0.045), and QTRT1 (HR 7.73, 95% CI 2.48-24.04, P<0.001) genes were significantly associated with TTFT in multivariable analyses when combined with the prognostic factors of the basic model, ultimately increasing the Harrell's c-index and the explained variation to 78.6% (versus 76.5% of the basic prognostic model) and 52.6% (versus 42.2% of the basic prognostic model), respectively. Also, the goodness of model fit was enhanced (χ2 = 20.1, P=0.002), indicating its improved performance above the basic prognostic model. In conclusion, DSAF-GS identified a group of significant genes for CLL prognosis, suggesting future directions for bio-molecular research.

INTERPRETING GENERATIVE ADVERSARIAL NETWORKS TO INFER NATURAL SELECTION FROM GENETIC DATA

Understanding natural selection in humans and other species is a major focus for the use of machine learning in population genetics. Existing methods rely on computationally intensive simulated training data. Unlike efficient neutral coalescent simulations for demographic inference, realistic simulations of selection typically requires slow forward simulations. Because there are many possible modes of selection, a high dimensional parameter space must be explored, with no guarantee that the simulated models are close to the real processes. Mismatches between simulated training data and real test data can lead to incorrect inference. Finally, it is difficult to interpret trained neural networks, leading to a lack of understanding about what features contribute to classification. Here we develop a new approach to detect selection that requires relatively few selection simulations during training. We use a Generative Adversarial Network (GAN) trained to simulate realistic neutral data. The resulting GAN consists of a generator (fitted demographic model) and a discriminator (convolutional neural network). For a genomic region, the discriminator predicts whether it is "real" or "fake" in the sense that it could have been simulated by the generator. As the "real" training data includes regions that experienced selection and the generator cannot produce such regions, regions with a high probability of being real are likely to have experienced selection. To further incentivize this behavior, we "fine-tune" the discriminator with a small number of selection simulations. We show that this approach has high power to detect selection in simulations, and that it finds regions under selection identified by state-of-the art population genetic methods in three human populations. Finally, we show how to interpret the trained networks by clustering hidden units of the discriminator based on their correlation patterns with known summary statistics. In summary, our approach is a novel, efficient, and powerful way to use machine learning to detect natural selection.

Timesweeper: accurately identifying selective sweeps using population genomic time series

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 developed to tackle this task, few are designed to leverage the potential of genomic time-series data. This is because in most population genetic studies of natural populations, only a single period of time can be sampled. 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. Serial sampling of organisms with shorter generation times has also become more feasible due to improvements in the cost and throughput of sequencing. 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 analyzes population genomic time-series data by first simulating training data under a demographic model appropriate for the data of interest, training a one-dimensional convolutional neural network on said simulations, and inferring which polymorphisms in this serialized data set were the direct target of a completed or ongoing selective sweep. We show that Timesweeper is accurate under multiple simulated demographic and sampling scenarios, identifies selected variants with high resolution, and estimates selection coefficients more accurately than existing methods. 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, as well as experimentally evolved populations where time-series data are often generated. 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 as a Python package for use by the community.

Uncovering Footprints of Natural Selection Through Spectral Analysis of Genomic Summary Statistics

Natural selection leaves a spatial pattern along the genome, with a haplotype distribution distortion near the selected locus that fades with distance. Evaluating the spatial signal of a population-genetic summary statistic across the genome allows for patterns of natural selection to be distinguished from neutrality. Considering the genomic spatial distribution of multiple summary statistics is expected to aid in uncovering subtle signatures of selection. In recent years, numerous methods have been devised that consider genomic spatial distributions across summary statistics, utilizing both classical machine learning and deep learning architectures. However, better predictions may be attainable by improving the way in which features are extracted from these summary statistics. We apply wavelet transform, multitaper spectral analysis, and S-transform to summary statistic arrays to achieve this goal. Each analysis method converts one-dimensional summary statistic arrays to two-dimensional images of spectral analysis, allowing simultaneous temporal and spectral assessment. We feed these images into convolutional neural networks and consider combining models using ensemble stacking. Our modeling framework achieves high accuracy and power across a diverse set of evolutionary settings, including population size changes and test sets of varying sweep strength, softness, and timing. A scan of central European whole-genome sequences recapitulated well-established sweep candidates and predicted novel cancer-associated genes as sweeps with high support. Given that this modeling framework is also robust to missing genomic segments, we believe that it will represent a welcome addition to the population-genomic toolkit for learning about adaptive processes from genomic data.

Genome-wide scans for selective sweeps using convolutional neural networks

MOTIVATION

Recent methods for selective sweep detection cast the problem as a classification task and use summary statistics as features to capture region characteristics that are indicative of a selective sweep, thereby being sensitive to confounding factors. Furthermore, they are not designed to perform whole-genome scans or to estimate the extent of the genomic region that was affected by positive selection; both are required for identifying candidate genes and the time and strength of selection.

RESULTS

We present ASDEC (https://github.com/pephco/ASDEC), a neural-network-based framework that can scan whole genomes for selective sweeps. ASDEC achieves similar classification performance to other convolutional neural network-based classifiers that rely on summary statistics, but it is trained 10× faster and classifies genomic regions 5× faster by inferring region characteristics from the raw sequence data directly. Deploying ASDEC for genomic scans achieved up to 15.2× higher sensitivity, 19.4× higher success rates, and 4× higher detection accuracy than state-of-the-art methods. We used ASDEC to scan human chromosome 1 of the Yoruba population (1000Genomes project), identifying nine known candidate genes.

Dispersal inference from population genetic variation using a convolutional neural network

The geographic nature of biological dispersal shapes patterns of genetic variation over landscapes, making it possible to infer properties of dispersal from genetic variation data. Here, we present an inference tool that uses geographically distributed genotype data in combination with a convolutional neural network to estimate a critical population parameter: the mean per-generation dispersal distance. Using extensive simulation, we show that our deep learning approach is competitive with or outperforms state-of-the-art methods, particularly at small sample sizes. In addition, we evaluate varying nuisance parameters during training-including population density, demographic history, habitat size, and sampling area-and show that this strategy is effective for estimating dispersal distance when other model parameters are unknown. Whereas competing methods depend on information about local population density or accurate inference of identity-by-descent tracts, our method uses only single-nucleotide-polymorphism data and the spatial scale of sampling as input. Strikingly, and unlike other methods, our method does not use the geographic coordinates of the genotyped individuals. These features make our method, which we call "disperseNN," a potentially valuable new tool for estimating dispersal distance in nonmodel systems with whole genome data or reduced representation data. We apply disperseNN to 12 different species with publicly available data, yielding reasonable estimates for most species. Importantly, our method estimated consistently larger dispersal distances than mark-recapture calculations in the same species, which may be due to the limited geographic sampling area covered by some mark-recapture studies. Thus genetic tools like ours complement direct methods for improving our understanding of dispersal.