AI-Mediated Reconstruction of Dinosaur DNA from Fossil Morphology and Extant Genomes

This work introduces a general three-step AI pipeline that can be applied to any extinct species, using only fossils and the morphology and genomes of living relatives. This machine learning pipeline is a skeleton-to-DNA framework that results in plausible probabilistic inferences about the genomes of the extinct species in question. The approach sits at the intersection of paleogenomics and de-extinction biology, and here is applied to dinosaurs. Genome to Skeleton: Train models that map whole-genome features to skeletal morphology using living species in the relevant clade (e.g., birds + non-avian reptiles for dinosaurs). This forward direction respects causality and establishes priors linking genotype to osteology. Skeleton to Genome: The system from step one is then inverted and trained on an additional set of living species. This inversion teaches the system to receive skeletal measurements and use them to generate genomic features. Fossil to Genome: Apply the inverse model to fossil skeletons to obtain genetic sequence candidates that are consistent with and constrained by the morphology of the extinct organism.