Colossal's GenoMorph borrows from more than 70 bird species to pick about 2,300 dodo edit targets

The usual method needs volume. "Normally, scientists can compare many genomes to understand which genetic differences are responsible for certain traits," Colossal CEO Ben Lamm told TechCrunch ^[5]. He went on: "We don't have that luxury with ancient DNA from extinct species whose DNA is severely limited, and we can only get a small number of complete genomes." ^[6] In Colossal's own projects the shortfall is 100 to 1, with one dodo genome against 100 for the mammoth ^[3][18].

GenoMorph moves the comparison to other animals. According to Lamm, the model tracks which stretches of DNA switch on while different bird species grow their beaks and faces, then predicts which genetic switches may have given the dodo its look ^[9]. Lamm said building it took more than a year ^[8].

Lamm's name for the result is a "treasure map for the dodo genome" ^[10]. In TechCrunch's account of his pitch, GenoMorph spares the team from searching through a bunch of potential genetic differences and points to the ones that matter ^[10]. What it has produced so far is a list of approximately 2,300 targets for the dodo's cranial and facial shape, which Lamm described as targets "that we can now test in the lab" ^[11]. TechCrunch's report does not say how many differences the search started with, so the 2,300 cannot be read as a reduction ^[11].

The user here is Colossal's own research team ^[2], and what they do with the output is narrower than the pitch. GenoMorph does not edit genes. Lamm said: "Identifying a potential edit doesn't mean we immediately introduce it into an animal or assume it's safe." ^[16] The check is a separate system. It uses lipid nanoparticles to deliver gene-editing tools into developing bird embryos, so the team can see what a change does without waiting for birds to reach sexual maturity ^[12]. Colossal hopes next to prove in a lab what GenoMorph predicts ^[15].

Two answers sort any AI tool sold as a fix for scarce data. The first is how close the stand-in data is to the thing being predicted. The second is how fast a prediction can be checked. Close data with a fast check is a tool a team can start using. Distant data with a slow check is a research bet and should be budgeted as one. The two mixed cells fall between. In our view, a team asked to defend such a tool can fill in both answers from the vendor's own documents.

GenoMorph's stand-in data is other birds, and its fast check is the embryo system that skips the breeding-cycle wait ^[12]. How close that data really is cannot be known until the embryo tests report, and Colossal says those tests are still ahead ^[11][15].