The University of Oregon has developed an innovative AI model, cxt, that revolutionizes the field of population genetics by reconstructing human ancestry from DNA. This cutting-edge technology, described in the Proceedings of the National Academy of Sciences, utilizes a modified GPT-2 architecture to analyze mutation patterns and estimate shared ancestry between genes. By learning from simulations of genetic evolution across various species, cxt translates mutation patterns into coalescence times, offering a novel approach to understanding our evolutionary history.
What sets cxt apart is its ability to handle large genomic datasets efficiently. It can infer pairwise coalescence curves for 50 haploid chromosomes in under five minutes on a single NVIDIA A100 GPU, outperforming traditional methods in terms of speed and accuracy. This is particularly useful for studying complex evolutionary scenarios, such as the evolution of insecticide resistance in malaria vectors.
In a real-world application, cxt was applied to human genomes from the 1000 Genomes Project and mosquito genomes from the Ag1000G consortium. It successfully recovered known patterns, such as the lactase persistence rise in the LCT region and the ancient variation preservation in the HLA region. The model's ability to infer deep genealogical structures and estimate ancient TMRCAs provides valuable insights into human and mosquito evolution.
The practical implications of this research are significant. By combining simulation-trained machine learning with traditional theory-driven methods, cxt enables researchers to process larger genomic datasets, handle messier sequence data, and study evolution more efficiently. While it may not replace the best theory-driven methods in every case, cxt's speed, flexibility, and adaptability make it a valuable tool for addressing complex questions in population genetics.
Looking ahead, the team aims to expand cxt's capabilities by reconstructing fuller genealogical trees, bringing it closer to the ancestral recombination graphs that population geneticists strive to recover. This development promises to further enhance our understanding of human ancestry and the intricate web of life's evolutionary history.
