Colossal 推出 GenoMorph AI，識別渡渡鳥基因編輯目標

Colossal Biosciences on October 14, 2026 unveiled GenoMorph, an artificial-intelligence platform developed to identify gene-editing targets for species with limited genomic data, and reported that the system has identified an estimated 2,300 candidate regulatory DNA elements associated with avian beak size and shape in the company’s dodo de-extinction program.

The Dallas-based company announced the platform alongside a set of advances across its avian pipeline: a substantially improved dodo genome assembly built with technology from its dire wolf project, an expanded Nicobar pigeon colony, newly derived Nicobar pigeon primordial germ cells, the first hatched chicken embryonic-stem-cell chimeras, and a lipid-nanoparticle system for delivering gene-editing tools into bird embryos. Colossal said the capabilities together provide the genomic, cellular and reproductive foundation needed to translate predicted dodo traits into experimentally validated edits. The announcement builds on advances the company reported in 2025 in pigeon primordial germ cells, gene-edited chicken surrogates and comparative avian genomes. Colossal, founded by entrepreneur Ben Lamm and geneticist George Church, describes itself as the first company to apply CRISPR technology to species de-extinction.

The GenoMorph Platform

GenoMorph combines comparative genomics, developmental biology and high-resolution 3D morphology to predict which regulatory elements may be active during development and to identify those most strongly associated with differences in physical form. Unlike protein-coding genes, which carry instructions for building an animal, regulatory DNA controls when, where and how strongly those instructions are activated, shaping traits such as the size and structure of a bird’s skull and beak.

To build the platform, Colossal’s avian team created what it calls an avian regulatory atlas from hundreds of samples collected from chickens, quail, ducks and pigeons, covering six stages of embryonic development and three regions of the developing face. The team combined those data with genomes and CT-derived skull measurements from more than 70 bird species, including the Nicobar pigeon, the dodo’s closest living relative, mapped using 541 anatomical landmarks.

Applied to the dodo program, the analysis identified 653 candidate regulatory elements associated with beak size and 1,698 associated with beak thickness and breadth. Colossal said the findings do not establish that the elements cause those traits but instead provide prioritized targets for experimental testing. Elements confirmed to influence craniofacial development may then be considered for precision genome editing in the Nicobar pigeon.

“De-extinction is as much a data problem as a biology problem,” said Ben Lamm, Colossal’s CEO and co-founder. He said that because so few dodo genomes survive, the company cannot rely on the population-level comparisons available for many living species, and that GenoMorph instead learns how regulatory DNA shapes living birds before identifying the genetic switches that may have contributed to the dodo’s distinctive anatomy. Lamm described the platform as a repeatable approach for extinct species whose genomic records are similarly limited.

From Prediction to Experimental Validation

Candidate elements are entering functional-validation studies, and those confirmed to influence craniofacial development may be prioritized for precision editing in the Nicobar pigeon genome, the company said. Chief Science Officer Beth Shapiro said the dodo’s defining anatomy emerged from networks of regulatory instructions that are difficult to interpret from ancient DNA alone, and that using the breadth of living bird diversity combined with AI allows GenoMorph to prioritize the regulatory changes most likely to have shaped the dodo and move them into functional validation.

“You cannot train AI to answer a precise developmental question using just generic data,” said Anna Keyte, Colossal’s avian species director. She said the team needed to capture how regulatory activity changes across species, developmental stages and regions of the face, and that building the dataset required more than a year of focused experimental work.

Preliminary findings have been presented at the Craniofacial Gordon Research Conference and the 2026 Society for Molecular Biology and Evolution annual meeting. A conference poster authored by Colossal researchers and academic collaborators details the underlying method: a convolutional neural network trained on chromatin-accessibility data from craniofacial dissections of pigeon, chicken, quail and duck embryos at six developmental stages, then used to predict accessibility across a 363-species bird alignment. The poster reports an average prediction accuracy, measured as area under the receiver-operating curve, of 0.863 within species and 0.848 across species. A phylogenetic regression between predicted accessibility and beak-length measurements drawn from VertNet across the 363 species found 210 of 293,686 tested open chromatin regions significantly associated with beak length at a false-discovery rate below 0.2, with the gene CACHD1 highlighted as a top candidate. The poster also reports that CACHD1 expression was elevated in long-beaked species throughout development of the frontonasal process, and that a separate transcriptomic analysis identified 356 genes with species-divergent expression dynamics between long- and short-beaked species.

A More Complete Dodo Genome

Colossal also reported progress on the dodo genome itself. Ancient DNA survives in short, damaged fragments, and conventional assembly methods map those fragments against a living reference genome, in this case the Nicobar pigeon, which can discard fragments too divergent to align. That can leave gaps in the assembly and bias the reconstructed genome toward the living species.

The company’s published iterative assembly method, developed through its dire wolf program, uses newly mapped dodo DNA to progressively update the reference genome, allowing previously unmatched fragments to fill gaps with each successive round until the number of newly recovered sequences reaches a plateau. Colossal reported that its dodo genome now has 95% of bases covered at least 5-fold, with an average coverage of 65-fold. Shapiro said the team is in the final stages of analyzing the dodo genome, with more to come after the genome paper goes through peer review.

Colony Growth, Stem Cells and Embryo Editing

Colossal expanded its Texas colony of Nicobar pigeons to 36 birds following successful breeding among established pairs. Eggs collected during the most recent breeding season were used to derive Nicobar pigeon primordial germ cells, the cells that eventually develop into sperm and eggs. Colossal said establishing these cells is essential to creating heritable genome edits and producing future generations of birds carrying selected dodo traits. The company is also developing a telomere-to-telomere Nicobar pigeon genome to improve on the existing chromosome-level assembly, and plans to remap newly available sequence data from the Rodrigues solitaire, an extinct relative of the dodo, against the improved reference to increase its coverage and completeness.

The company’s scientists also hatched the first chimeric chickens produced through a new platform built on avian embryonic stem cells, which Colossal said confirms that cultured cells retain the developmental potential to integrate into a host embryo and contribute to its tissues. The culture system is being expanded to pigeons so that edited stem cells can be tested in chimeric pigeons for their predicted effects on phenotype before any edits are made to primordial germ cells, establishing a phenotype-screening platform.

Finally, Colossal reported developing a lipid-nanoparticle delivery platform that introduces DNA, RNA and CRISPR gene-editing tools into developing bird embryos while improving embryo viability over traditional delivery methods. In the company’s experiments, editing the bone-development regulator Runx2 produced a clear craniofacial phenotype, and the system has been used in rock dove and duck embryos. Colossal said its scientists are using the platform to validate GenoMorph-identified regulatory elements in near real time, that the strongest candidates are being engineered into Nicobar pigeon primordial germ cells for further precision-editing studies, and that a manuscript describing the platform has been submitted to the bioRxiv preprint server.