University of Oregon researchers develop AI model to trace ancestral lineages from DNA
Updated
Updated · BIOENGINEER.ORG · May 5
University of Oregon researchers develop AI model to trace ancestral lineages from DNA
9 articles · Updated · BIOENGINEER.ORG · May 5
Published in PNAS, the GPT-2-based system was retrained on simulated genomes from bacteria, rodents, mosquitoes and primates to estimate when gene pairs last shared a common ancestor.
Researchers said it cuts analyses from hours or days to minutes, including for a mosquito chromosome, while handling incomplete genomic data that can hinder classical statistical methods.
The tool could help date the emergence of insecticide resistance and other traits, and the team aims to extend it from pairwise comparisons to full genealogical trees.
Can this AI predict the next malaria super-mosquito, or just explain how the last one evolved?
If AI misreads DNA's 'grammar,' can we trust it to predict the next superbug threat?
Decoding Evolutionary History at Scale: cxt’s Breakthrough in Fast, Accurate Coalescence Time Estimation
Overview
In June 2025, the University of Oregon introduced cxt, a novel AI model that transforms evolutionary inference by treating mutation patterns as a biological language translated into ancestral histories. Built on a decoder-only transformer architecture and trained on diverse synthetic genetic data, cxt matches the accuracy of traditional methods while delivering over a million coalescence time predictions in minutes. Its well-calibrated uncertainty estimates and robust generalization enable applications from tracking malaria mosquito resistance to uncovering human evolutionary adaptations. Experts hail cxt's speed and scalability but emphasize the need for explainable AI to address potential biases arising from synthetic training data and skewed genomic databases.