The dodo project has a data problem before it has an editing problem. Colossal Biosciences says it has one high-quality dodo genome, while its mammoth work can draw on around one hundred genomes. One specimen makes it difficult to separate a genetic feature unique to the dodo from ordinary variation that might have existed within the species.[2][3]
GenoMorph takes a different route. It learns from living birds: developmental samples from chickens, quail, ducks and pigeons; genomes and CT-derived skull measurements from more than 70 species; and 541 anatomical landmarks. The system then ranks regulatory DNA associated with differences in beak size, thickness and breadth.[1]
The announced shortlist
| Measure | Value | Context |
|---|---|---|
| Beak-size candidates | 653 regulatory elements | Associated with size across the comparative analysis.Source 1 |
| Thickness and breadth candidates | 1,698 regulatory elements | Associated with other dimensions of beak form.Source 1 |
| Total candidates | 2,351 | The company rounds the public figure to approximately 2,300. |
| Living bird species | More than 70 | Used for comparative genomic and skull-shape evidence.Source 1Source 2 |
| Nicobar pigeon colony | 36 birds | The closest living relative is central to reproductive and developmental work.Source 2 |
A regulatory switch is not a beak gene
Many visible traits do not arise from one protein-coding gene. Regulatory elements influence when, where and how strongly other genes are expressed. A sequence can matter only during a particular embryonic stage or in one region of the developing face. The same gene may contribute to several structures depending on those controls.
This is why the training data included six embryonic stages and three facial regions. GenoMorph is trying to connect changing gene activity with changing anatomy across related birds. The model can nominate a sequence that repeatedly tracks a shape difference. It cannot, from correlation alone, show that editing that sequence will reproduce the intended trait without other effects.
From a prediction to a viable inherited trait
| Stage | Question | Evidence still needed |
|---|---|---|
| Candidate ranking | Which DNA elements are associated with the target morphology? | Robust comparative and developmental signal.Source 1 |
| Functional validation | Does changing the element alter gene expression or tissue development? | Cell, embryo and molecular assays.Source 2 |
| Precision editing | Can the change be introduced without damaging other regions? | Sequence verification and off-target analysis. |
| Germ-line transmission | Can the edited change pass through sperm or eggs? | Stable avian germ-cell and breeding evidence.Source 2Source 3 |
| Whole-animal development | Does the combination produce a healthy bird with the intended anatomy? | Developmental, behavioural and welfare assessment. |
| Ecological release | Could a population live safely in a prepared habitat? | Regulation, community consultation and ecosystem work.Source 3 |
The bottleneck shifts from search to experimental throughput
A shortlist of 2,300 is dramatically smaller than the space of all sequence differences, but it is still too large to treat as an editing recipe. Colossal says it has developed lipid-nanoparticle delivery into bird embryos so candidate changes can be tested without waiting for every bird to reach sexual maturity. That could accelerate validation while making the quality of each assay more important.[2]
The company has also reported progress culturing avian primordial germ cells—the precursors of sperm and eggs—and expanding its Nicobar pigeon colony. Those reproductive tools are separate from GenoMorph’s predictions. De-extinction requires them to converge: a good candidate, a reliable edit, a heritable route and a healthy developmental outcome.[2][3]
The most useful way to read the announcement is not “AI found the dodo code.” It found a tractable order for experiments. In a project with scarce ancient DNA, that is a real contribution—but the organism will be decided by what survives the laboratory, the embryo and the ecosystem, not by the ranking score.
Sources & further reading
- Colossal Accelerates the Full Dodo De-extinction Pipeline Colossal Biosciences via PR Newswire · 14 Oct 2026 · Accessed 17 Oct 2026
- Colossal Biosciences announces new AI platform to help accelerate de-extinction efforts TechCrunch · 14 Oct 2026 · Accessed 17 Oct 2026
- US company announces progress toward resurrecting the extinct dodo Reuters · 14 Oct 2026 · Accessed 17 Oct 2026