Somite Therapeutics’ founders (L to R): Allon Klein, Cliff Tabin, Olivier Pourquie, Jonathan ... [+]
Five scientists walked into a lab late last year and decided to establish a startup combining their wide-ranging expertise, practical experience and breakthrough innovations. The goal of their startup, Somite.ai, is to improve the lives of millions of people by leveraging AI to produce human tissue for cell therapies. The market for the therapies for the diseases they aim to address is estimated at $250 billion.
Somite.ai is a potent mix of state-of-the art stem cell biology, medicine, genetics, engineering, statistics, computer science, and machine learning. The co-founders are Olivier Pourquie, Professor of Genetics, Harvard Medical School and of Pathology, Brigham and Women’s Hospital; Allon Klein, Associate Professor of Systems Biology, Harvard Medical School; Cliff Tabin, Professor and Chair of Genetics, Harvard Medical School; Jonathan Rosenfeld, Head of the FundamentalAI group at MIT FutureTech; and Micha Breakstone—his PhD is in Cognitive Science, but he opted for a non-academic career, becoming a successful AI entrepreneur (selling Israeli startup Chorus.ai for $575 million, among other endeavors).
“We're developing life language models,” says Breakstone, “we're literally learning the game of life.” Many established biotech companies and startups today are trying to better understand cell development and establish new paths to cell replacement therapy or CRT. What makes Somite.ai stand out in this crowded market?
In addition to assembling a dream team, Breakstone points to three fundamental differentiators: the focus on somites; using embryo digital twins (a computational model representing the composition and organization of an embryo) to guide and control the discovery, optimization and production variability in CRT; and novel use of AI.
Somites are the embryonic structures that are responsible for producing the musculoskeletal system and related tissues. “No other lab today can produce Somites at our level of efficacy and accuracy,” says Breakstone. In 2015 and 2016, Pourquie published two seminal papers demonstrating how to develop these type of cells in an effective way.
Meanwhile, Klein and other researchers successfully integrated tools from molecular biology, engineering, statistics and computer science, to produce “movies” showing how, in the words of Science, “a single fertilized egg gives rise to the multitude of cell types, tissues, and organs that fit together to make a body.” Klein, who received the James Prize for this breakthrough, says: “This technology is essentially the foundation now for turning developmental biology, which has been an expert-driven science, into a data science.”
The new insights into cell differentiation and tissue development paved the way to today’s cell replacement therapy. “It’s the most sophisticated therapy that has ever been invented,” says Klein. Instead of therapies based on a drug delivered to a cell, now the treatment itself is “a living cell that can sense, respond, form tissues, regenerate.” CRT today, says Klein, represents enormous potential but also enormous complexity.
The key challenge for CRT is to find the step-by-step instruction set that leads to the desired cell types at large scale, with minimal contaminants of unwanted cell types, and with high reproducibility. The two main approaches for achieving this goal are “direct programming” (through a process known as gene editing) and “guided differentiation.”
The gene editing strategy, says Klein, is a “very exciting” vision, promising that “if you can just bring cells right to the end point, they will figure everything else out and they will become physiologically normal.” Somite.ai, however, adheres to the slower and safer guided differentiation approach, convinced that “every single step that the cell goes through in the process of development is important for shaping its final function and identity. And we need to take cells as closely as possible through these steps in order to make sure that we get to the end point.” The environment in which the cell grows “can matter a great deal,” says Klein.
The new, data-rich approach to understanding cell differentiation and tissue development, the creation of embryo digital twins and the accumulation of data generated by many labs have led to a data explosion. For example, data from single-cell RNA sequencing is currently doubling every six months.
This is where AI can make a big difference. “Data by itself is valuable, but we need to know how to use this data systematically. We also need to update the data continuously, to take what we learn in the dish and feed it back into a system which can learn on the fly,” says Klein. Self-supervised machine learning, trained on existing data troves and what’s produced in the lab, coupled with the knowledge contained in thousands of papers published in recent years mined by text-processing domain-specific large language models, facilitate automating and accelerating the discovery and optimization of the step-by-step guide to successful CRT.
Somite.ai announced last week that it has raised $5.3 million in pre-seed funding. The round was led by TechAviv, joined by Next Coast Ventures, Trust Ventures, Texas Venture Partners, Lerer Hippeau and others. The startup has already hired two translational scientists, experienced in the complex process of commercializing FDA-approved treatments. Targeting therapies that have the potential to cure a wide range of diseases that involve the loss or deficiency of cell populations such as muscular dystrophies, obesity and diabetes, Somite.ai plans to bring, within the next two years, its first therapeutic asset to phase-1 clinical trials.
“The future of medicine lies at the intersection of AI and biology,” says Breakstone. Successfully realizing this vision entirely depends on the quality of human intelligence involved, on melding together the best of multiple scientific disciplines, and on the art of seeing around corners.
