Somite AI Raises $47M Series A To Reinvent Cell Replacement Therapy

Forbes

13-05-2025

Regenerative medicine and therapeutic stem cell therapy to regrow damaged cells as treatment for ... More disease
Startup Somite AI announced today that it has raised over $47 million in a Series A funding round, bringing its total funding to about $60 million. The round was led by Khosla Ventures, with participation from SciFi VC, the Chan Zuckerberg Initiative, Fusion Fund, Ajinomoto Group Ventures, Pitango HealthTech, TechAviv, Harpoon Ventures, and prominent angel investors such as Fidji Simo, the new CEO of Applications at OpenAI.
Legendary investor Vinod Khosla said in a statement: 'Somite AI's foundation models, once fully developed and validated, will not only create value for their own pipeline, but have the potential to reshape the entire field of human cell therapy.'
Somite's initial ambition was to design a 'life language model' to drive the development of Somites, the embryonic structures at the origin of the musculoskeletal system, and use them in cell replacement therapy targeting particular ailments such as Duchenne muscular dystrophy, the most common hereditary neuromuscular disease. With today's announcement, Somite is publicly broadening its scope, aiming to revolutionize the production of any human cell type through its AI foundation model platform, DeltaStem, to address a broad spectrum of diseases, including metabolic disorders such as Type 1 Diabetes, orthopedic conditions, muscular diseases, and blood disorders.
When Somite was established just over a year ago, the founders asked themselves what they would be able to do if they could create any type of human cell. 'We'd be able to replace and replenish any diseased or damaged tissue in the body,' says co-founder and CEO Micha Breakstone, defining their mission as reverse engineering of stem cell biology with artificial intelligence.
'Today, we only know how to reliably produce about 10 types of cells, but there are over 5,000 types of cells in the human cell atlas,' says Breakstone. 'There are more possible ways of creating a specific cell than there are atoms in the universe.'
To take a stem cell to a specific mature state, researchers adjust both its environment and the specific biological signals it receives—such as the concentration of a growth factor or a change in pH—learning through step-by-step experimentation the precise sequence of conditions that guide its developmental trajectory. How do you re-engineer a process that takes many years, tens of millions of dollars, and lots of trial and error, with no guaranteed result?
As in other domains (e.g., quantum computing error correction), where the 'search space,' the initial set of potential solutions to a problem, is beyond human comprehension (e.g., larger than the number of atoms in the universe), AI could be of great help. However, you need vast quantities of data to successfully apply today's AI or deep learning.
The first ingredient in Somite's recipe for developing a cell foundation model is its innovative method for efficiently generating data at scale in a completely natural way. Professor Alon Klein, another Somite co-founder, had a thinking-inside-the-box breakthrough moment when he asked, as Breakstone tells it, 'What would happen if instead of taking the signals to the cells, we take the cells to the signals?'
The box or boxes in this case are semi-permeable capsules, each containing a few cells, with minuscule 'windows' allowing the signals to come in and affect the cells. The Somite lab team sends millions of these capsules through many signals or conditions, and each cell-condition encounter is recorded with a barcode. Put all the barcodes together and 'you get an address that correlates with exactly the specific trajectory that the capsule went through, a beautiful model of the development of the embryo,' explains Breakstone. This capsule technology generates cell state transition data at an unprecedented scale, achieving 1000x greater efficiency than existing methodologies.
For Jonathan Rosenfeld, another Somite co-founder, this data serves as the basis for developing an AI foundation model for the human cell. Currently, Somite is running experiments at roughly a one-million-condition scale, mapping the developmental paths embryonic cells take under these varied conditions. It plans to reach 10 million conditions by the end of the year.
Given the enormous size of the 'search space,' even 100 million would be just a drop in the vast ocean of 'protocols,' all the possible trajectories of cell development. 'However,' says Breakstone, 'it's a drop large enough to learn from, and that's the beauty of deep learning. From a relatively small amount of data, we can develop and train a working model to generate novel protocols that we haven't seen before.'
With its DeltaStem platform, Somite aims to overcome key challenges of current lab-based stem cell development—the prohibitively expensive, slow, manual, trial-and-error methods researchers use to guide stem cells toward specific mature cell types. Major obstacles include purity (percentage of desired cells produced), scalability (efficient cell manufacturing at therapeutic scale), and reproducibility (consistent results across batches). 'What AlphaFold did by predicting protein structures, transforming structural biology, DeltaStem aims to achieve for stem cell biology—predicting the conditions needed to precisely generate mature human cell types at scale,' says Breakstone.
Unlike biotech companies traditionally engaged in individual drug development, 'TechBio' companies like Somite focus on developing broad technology platforms addressing multiple indications. Their approach is data-driven, and their goal is to develop proprietary AI that reduces costs and risks, shortens drug and treatment development time, and accelerates the shift to more personalized medicine. The success of Moderna and the Nobel Prize awarded to AlphaFold have attracted increased venture capital and corporate investment in applying data and AI to current healthcare challenges, including improving regenerative medicine: 'Artificial intelligence presents broad utility in the discovery and development of new biotherapeutics.'
Somite's AI foundation model and platform will help build a future 'where we can create any human cell on demand, like a supply of cellular spare parts to repair or replace diseased or damaged tissue,' says Breakstone.