SignificanceDate palm (Phoenix dactylifera L.) is one of the oldest tree crop species in the world and is a major fruit crop of arid regions of the Middle East and North Africa. We use whole-genome sequence data from a large sample of P. dactylifera and its wild relatives to show that hybridization between date palms and Phoenix theophrasti Grueter—a species endemic to the Eastern Mediterranean—is associated with the diversification of date palm.

Detecting signals of selection from genomic data is a central problem in population genetics. Coupling the rich information in the ancestral recombination graph (ARG) with a powerful and scalable deep-learning framework, we developed a novel method to detect and quantify positive selection: Selection Inference using the Ancestral recombination graph (SIA). Built on a Long Short-Term Memory (LSTM) architecture, a particular type of a Recurrent Neural Network (RNN), SIA can be trained to explicitly infer a full range of selection coefficients, as well as the allele frequency trajectory and time of selection onset. We benchmarked SIA extensively on simulations under a European human demographic model, and found that it performs as well or better as some of the best available methods, including state-of-the-art machine-learning and ARG-based methods. In addition, we used SIA to estimate selection coefficients at several loci associated with human phenotypes of interest. SIA detected novel signals of selection particular to the European (CEU) population at the MC1R and ABCC11 loci. In addition, it recapitulated signals of selection at the LCT locus and several pigmentation-related genes. Finally, we reanalyzed polymorphism data of a collection of recently radiated southern capuchino seedeater taxa in the genus Sporophila to quantify the strength of selection and improved the power of our previous methods to detect partial soft sweeps. Overall, SIA uses deep learning to leverage the ARG and thereby provides new insight into how selective sweeps shape genomic diversity.

Insular organisms often evolve predictable phenotypes, like flightlessness, extreme body sizes, or increased melanin deposition. The evolutionary forces and molecular targets mediating these patterns remain mostly unknown. Here we study the Chestnut-bellied Monarch (Monarcha castaneiventris) from the Solomon Islands, a complex of closely related subspecies in the early stages of speciation. On the large island of Makira M. c. megarhynchus has a chestnut belly, whereas on the small satellite islands of Ugi, and Santa Ana and Santa Catalina (SA/SC) M. c. ugiensis is entirely iridescent blue-black (i.e., melanic). Melanism has likely evolved twice, as the Ugi and SA/SC populations were established independently. To investigate the genetic basis of melanism on each island we generated whole genome sequence data from all three populations. Non-synonymous mutations at the MC1R pigmentation gene are associated with melanism on SA/SC, while ASIP, an antagonistic ligand of MC1R, is associated with melanism on Ugi. Both genes show evidence of selective sweeps in traditional summary statistics and statistics derived from the ancestral recombination graph (ARG). Using the ARG in combination with machine learning, we inferred selection strength, timing of onset and allele frequency trajectories. MC1R shows evidence of a recent, strong, soft selective sweep. The region including ASIP shows more complex signatures; however, we find evidence for sweeps in mutations near ASIP, which are comparatively older than those on MC1R and have been under relatively strong selection. Overall, our study shows convergent melanism results from selective sweeps at independent molecular targets, evolving in taxa where coloration likely mediates reproductive isolation with the neighboring chestnut-bellied subspecies.

Detecting signals of selection from genomic data is a central problem in population genetics. Coupling the rich information in the ancestral recombination graph (ARG) with a powerful and scalable deep learning framework, we developed a novel method to detect and quantify positive selection: Selection Inference using the Ancestral recombination graph (SIA). Built on a Long Short-Term Memory (LSTM) architecture, a particular type of a Recurrent Neural Network (RNN), SIA can be trained to explicitly infer a full range of selection coefficients, as well as the allele frequency trajectory and time of selection onset. We benchmarked SIA extensively on simulations under a European human demographic model, and found that it performs as well or better as some of the best available methods, including state-of-the-art machine-learning and ARG-based methods. In addition, we used SIA to estimate selection coefficients at several loci associated with human phenotypes of interest. SIA detected novel signals of selection particular to the European (CEU) population at the MC1R and ABCC11 loci. In addition, it recapitulated signals of selection at the LCT locus and several pigmentation-related genes. Finally, we reanalyzed polymorphism data of a collection of recently radiated southern capuchino seedeater taxa in the genus Sporophila to quantify the strength of selection and improved the power of our previous methods to detect partial soft sweeps. Overall, SIA uses deep learning to leverage the ARG and thereby provides new insight into how selective sweeps shape genomic diversity.

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 thedomain-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.