Colossal's GenoMorph AI surfaces 2,300 gene edit targets for dodo de-extinction

Colossal Biosciences has unveiled an AI platform called GenoMorph that it says will accelerate its de-extinction programs by pinpointing which genes are worth editing when only fragments of ancient DNA are available. According to TechCrunch, which first reported the news, the company is already applying the tool to its effort to bring back the dodo.

What GenoMorph does

The platform combines comparative genomics, 3D morphology and developmental biology to flag candidate genes in species where genomic data is scarce. That scarcity is the core problem for the dodo project: Colossal is working from a single dodo genome, compared with roughly 100 genomes available to its woolly mammoth effort. The company told TechCrunch the gap comes down to climate, since cold environments preserved mammoth DNA far better than the tropical conditions where the dodo lived.

Colossal CEO and co-founder Ben Lamm explained to TechCrunch that scientists normally compare many genomes to work out which genetic differences drive particular traits, but that approach collapses when only a small number of complete genomes exist for an extinct species.

How it was built

To train the system, Colossal gathered developmental data from chickens, quail, ducks and pigeons, and paired it with genetic information and 3D skull measurements from more than 70 bird species. Lamm said the effort took over a year to build. The AI learns which DNA regions are active as birds develop their beaks and facial structures, then uses that knowledge to predict which genetic switches may have produced the dodo's distinctive appearance. Lamm likened the platform to a "treasure map" for the dodo genome, narrowing the search to the genetic differences that actually matter.

Targets identified, lab tests next

GenoMorph has so far identified approximately 2,300 promising DNA editing targets for dodo craniofacial morphology that the team can now test in the lab, Lamm told TechCrunch. He framed the tool as a way past one of de-extinction's biggest bottlenecks: working out which changes make an extinct species unique in the first place.

Identifying an edit is not the same as proving it works, however. Colossal has also developed a delivery system that uses lipid nanoparticles to carry gene-editing tools directly into developing bird embryos, letting researchers observe how a genetic change affects development without waiting through a full breeding cycle for birds to reach sexual maturity.

Supporting progress on pigeons

Alongside the platform launch, the company said its colony of Nicobar pigeons, the dodo's closest living relative, has grown to 36 birds, and that it has successfully created the essential cells that go on to become eggs and sperm. Lamm called those cells a critical component of the strategy for creating birds that carry dodo traits. The immediate next step is proving GenoMorph's predictions in the laboratory.

On safety, Lamm emphasized that GenoMorph does not edit genes itself; it shortlists changes worth investigating. Identifying a candidate edit, he said, does not mean it is immediately introduced into an animal or assumed to be safe, and more precise identification makes downstream decisions better informed.

In a related development reported by TechCrunch, XPRIZE founder and Colossal investor Peter Diamandis announced a new venture firm, NeoGenesis Capital, to back companies building on Colossal's programmable biology platform across health, agriculture and conservation.

Why it matters

De-extinction has always been constrained less by gene-editing tools than by knowing where to point them. GenoMorph is an attempt to convert a data problem, scarce ancient DNA, into a machine-learning problem, using information from living relatives to infer what fragmentary genomes cannot show directly. If the predicted dodo edits hold up in lab tests, the same approach could transfer to other species with limited genomic records, which describes most extinct animals. Lamm also suggested the underlying technologies are converging in ways that could support projects elsewhere on the tree of life. The announcement also shows AI moving beyond pattern-matching in text and images toward generating testable biological hypotheses, with the lab, not the model, remaining the final arbiter.