Coalescence and translation: A language model for population genetics.

Probabilistic models such as the sequentially Markovian coalescent have long provided a powerful framework for population genetic inference, enabling reconstruction of demographic history and ancestral relationships from genomic data. However, these methods are inherently specialized, relying on predefined assumptions and/or limited scalability. Recent advances in simulation and deep learning provide an alternative approach: learning directly to generalize from synthetic genetic data to infer hidden evolutionary processes. Here we reframe the inference of coalescence times as a problem of translation between two biological languages: the sparse, observable patterns of mutation along the genome and the unobservable ancestral recombination graph that gave rise to them. Inspired by large language models, we develop cxt, a decoder-only transformer that autoregressively predicts coalescent events conditioned on local mutational context. We show that cxt performs competitively with state-of-the-art Markov Chain Monte Carlo-based likelihood models across a broad range of demographic scenarios, matching their accuracy in-distribution and approaching it in out-of-distribution, with the potential for improvement via fine-tuning. Trained on simulations spanning the stdpopsim catalog [J. R. Adrion et al., eLife 9, e54967 (2020); M. E. Lauterbur et al., eLife 12, e84874 (2023); G. Gower et al., Accessible, realistic genome simulation with selection using stdpopsim. bioRxiv [Preprint] (2025). https://doi.org/10.1101/2025.03.23.644823.], the model generalizes robustly and enables efficient inference at scale, producing over a million coalescence predictions in minutes. cxt also produces well-calibrated approximate posteriors, enabling principled uncertainty quantification. We apply cxt to population genomic data from both humans and mosquitoes, highlighting the model's ability to deal with the complexities of empirical data.