- What happened: Colossal Biosciences announced an AI platform called GenoMorph and reported that it identified about 2,300 candidate DNA edits linked to how the extinct dodo's skull and face formed.
- Why it matters: Even for an extinct species with little surviving genomic data, comparing developmental data and skull shapes from living birds may help narrow down the DNA regions tied to the traits researchers want to recreate.
- What to watch next: How many of the AI-selected candidates actually produce the intended physical changes, and whether edits made through gene editing can be passed to offspring through germ cells.
On October 14, 2026, U.S. biotechnology company Colossal Biosciences announced GenoMorph, an AI platform that searches for DNA regions that could be targeted by gene editing to recreate traits of the extinct dodo. Co-founder and CEO Ben Lamm explained how it works in an exclusive TechCrunch interview. According to the company, it identified about 2,300 promising DNA edit candidates related to the shape of the dodo's skull and face.
GenoMorph tries to make up for the extremely limited genomic information available on the dodo by combining developmental data from living birds with 3D skull shapes. But identifying edit candidates is not the same as actually recreating dodo traits. The next steps are to test in embryos whether altering these candidate DNA regions causes the intended physical changes, and to establish technology for passing those traits on to offspring.
Only one dodo genome is available; data from living birds fills the gap
Colossal says it currently has just one dodo genome available for its research. By contrast, its mammoth revival project has access to about 100 genomes.
According to Lamm, DNA from mammoths, which lived in cold regions, tends to be better preserved, while collecting comparable DNA material from the tropical dodo is much harder. These figures compare the number of genomes available to Colossal's research, not the total number of genomes sequenced worldwide.
Being able to compare the genomes of many individuals makes it easier to find DNA differences associated with particular physical traits. But for an extinct species like the dodo, with few available genomes, it is difficult to judge whether a sequence difference is actually what produces the species' distinctive form.
Reading a DNA sequence and understanding how that sequence shapes an organism's appearance are separate challenges.
GenoMorph tries to solve this problem by comparing the dodo with living birds.
Colossal says it collected developmental data from chickens, quail, ducks, pigeons and other birds, and combined genetic information from more than 70 bird species with 3D measurements of their skulls.
As Lamm describes it, the AI learns which DNA regions are active as a bird's beak and face take shape. Based on that, it predicts regulatory DNA regions, which control how genes function, that may have been involved in the dodo's distinctive form.
In other words, rather than learning only from the dodo's DNA sequence, it draws on the developmental processes observable in living birds and on differences in form between species.
The approach is meant to narrow a limited pool of genetic information from an extinct species down to DNA regions worth examining in practice.
The figure of about 2,300 should be understood in that light. What Colossal reports having identified are DNA edit candidates considered promising for recreating the shape of the dodo's skull and face; it does not mean that about 2,300 genes were actually modified.
Experiments will be needed to determine what proportion of these candidates actually cause the intended physical changes.
Beak shape changes with where and when genes are active
It has long been studied how bird face and beak shapes vary with how genes act during development.
A classic example is the study of Darwin's finch beaks published in the journal Nature in 2006 by Arhat Abzhanov and colleagues.
The research team investigated why beak shapes differ among bird species and focused on a signaling pathway involving a molecule called calmodulin.
According to the paper's public abstract, calmodulin expression was higher in cactus finches, which have long, pointed beaks. When the researchers artificially activated the same pathway in part of the face of chicken embryos, the upper beak lengthened.
This is an example of testing a relationship between gene activity and form, found by comparing different birds, through an experiment intervening in the embryo of another bird.
What this research shows is that identifying the genes involved is not enough to understand the mechanisms that produce a distinctive form.
It is necessary to clarify what kind of gene activity occurs, in which part of the body, and at which stage of development.
GenoMorph's combination of developmental data and skull morphology, rather than genetic information alone, appears aimed at finding such relationships across a broad range of birds.
However, it cannot be confirmed from the reporting whether the roughly 2,300 candidates identified include DNA regions related to calmodulin.
Colossal has long said, in its official dodo revival plan, that it would use comparative genomic analysis and machine learning to predict physical traits from genetic information and apply that to gene editing.
What is new in this announcement is that GenoMorph, the platform that puts that policy into practice, the data used to train it, and the scale of the roughly 2,300 edit candidates have been disclosed. It is not an announcement of a new discovery about how bird beaks form.
From finding edit candidates to passing dodo traits to offspring
Colossal is also refining gene-editing technology to test the candidates identified by GenoMorph in practice.
According to the announcement, the company has developed a method that uses lipid nanoparticles to deliver the molecules needed for gene editing directly into developing bird embryos.
Lamm explains that this method makes it possible to examine how genetic modifications affect development and body shape without waiting for a bird to grow up and become able to breed.
This could shorten the time needed to test the DNA edit candidates GenoMorph selects in actual embryos.
However, even if a specific physical change can be induced in an embryo, that alone does not make it possible to propagate birds with that trait through breeding.
To pass gene-edited changes to offspring, it must be confirmed that the right genetic information is built into the germ cells that produce eggs and sperm and is transmitted to the next generation.
Organizing Colossal's efforts based on TechCrunch's reporting, a peer-reviewed paper published in 2026 and the official dodo revival plan, the research can be divided into three broad stages.
| Research stage | Results published so far | Challenges still to be tested |
|---|---|---|
| Selecting DNA edit candidates | Announced that GenoMorph identified about 2,300 candidates related to the shape of the dodo's skull and face | Whether each candidate causes the predicted physical change |
| Testing physical changes in embryos | Described developing technology to use lipid nanoparticles to deliver gene-editing molecules into bird embryos | Editing efficiency, effects on the intended form, embryo survival and health effects |
| Passing genetic information to the next generation | Cultured primordial germ cells from the rock pigeon and confirmed that they migrate to the gonads of transplanted embryos | Whether they form properly functioning eggs and sperm and can pass edited genetic information to offspring |
This table does not compare the performance of each technology under the same conditions; it organizes the results published so far by research stage.
In particular, the technology for delivering gene-editing molecules into embryos and the technology for culturing and transplanting germ cells differ in both purpose and experimental subject. Being able to check physical changes quickly in embryos does not mean the technology for passing those traits to the next generation has been completed.
Pigeon germ cells cultured for more than 120 days and shown to migrate to the embryonic gonads
Relevant to the challenge of passing genetic information to offspring is research published in the Journal of Heredity by Martin W. Nicholson and colleagues.
The team, which included Colossal researchers, collected primordial germ cells, the precursors of eggs and sperm, from rock pigeons and succeeded in culturing them for more than 120 days.
They also labeled the cultured cells and injected them into rock pigeon and chicken embryos. Four days after injection, they confirmed that the transplanted cells had migrated to the embryos' gonads.
This result shows that avian germ cells can not only be cultured over a long period but also retain, after culture, their innate tendency to travel to their natural destination in the embryo. It is an important technical advance for genetic modification of birds and for preserving genetic resources.
However, a germ cell reaching the gonad is not the same as that cell becoming a normal egg or sperm able to pass genetic information to offspring.
The paper's authors also describe whether the cultured primordial germ cells actually have the ability to pass genetic information to the next generation as an issue for future evaluation.
Colossal also announced this time that its captive population of Nicobar pigeons, the dodo's close relatives, has grown to 36 birds and that it has succeeded in producing cells that are precursors of eggs and sperm.
The company's official plan states that it intends to use Nicobar pigeon cells and genomes as the foundation for genetic modifications to recreate dodo traits.
However, this announcement must be distinguished from the results of the peer-reviewed paper using rock pigeon primordial germ cells. Confirming cell culture and migration to the gonads in a different bird does not mean breeding has succeeded in Nicobar pigeons with the same technology.
To judge how much AI speeds research, experimental success rates matter more than candidate counts
The figure of about 2,300 presented in the GenoMorph announcement suggests that AI may be able to find large numbers of gene-editing candidates.
But having many candidates does not necessarily mean research has actually become more efficient.
The TechCrunch article does not give specific figures from which to judge GenoMorph's prediction accuracy or how much it has shortened research time compared with conventional methods.
The roughly 2,300 is the result of narrowing the targets to examine in future experiments, and the number itself does not indicate the technology's success rate.
To evaluate how much AI accelerates research, it is necessary to confirm how many of the candidates actually tested caused the predicted physical changes.
Even if the intended physical change can be achieved, questions remain about whether the embryo develops normally and can grow into a healthy individual. Confirming reproducibility, meaning that the same results are obtained under different experimental conditions, is also essential.
Ultimately, if the goal is to maintain birds with dodo traits as a breeding population, it must also be shown that the edited genetic information is passed on to offspring through germ cells.
Only after these stages can it be assessed whether the DNA edit candidates chosen by the AI were actually effective in recreating traits of an extinct species.
On the other hand, if AI can judge which DNA regions to examine first in research on extinct species with limited genomic information, it has a significant role to play.
That is because limited time and experimental resources can be concentrated on candidates with a higher chance of success. What matters is not only the ability to generate large numbers of candidates but whether it can accurately pick out the ones worth testing.
Colossal has outlined plans to use its technology for transplanting pigeon primordial germ cells into chicken hosts not only in research to recreate traits of extinct species but also in the conservation of living bird species.
If GenoMorph's morphological predictions lead to verification in embryos and reliable transmission of genetic information to the next generation, they could be applied both to research recreating traits of extinct species and to technology for preserving the genetic resources of living birds.
The focus going forward will be how many of the roughly 2,300 edit candidates can actually produce the intended physical changes, and whether those traits can be passed to healthy offspring. Whether AI predictions can be shown to translate into real biological form and inheritance will be a key yardstick for judging GenoMorph's technical value.