Researchers at Harvard are pushing the boundaries of biology by using advanced computation to understand how cells organize themselves into tissues and organs. Their work could ultimately transform organ design and cellular reprogramming, allowing scientists to move from trial-and-error experimentation to predictable, computer-guided design.
Turning Biology Into an Optimization Problem
At its core, this effort treats one of the most fundamental processes in biology as a solvable problem. Cells naturally self-organize, dividing and forming structures like organs, limbs, and tissues. For decades, scientists have tried to harness this process to build artificial organs or better understand diseases such as cancer. But the challenge has always been that engineering cells to achieve a desired outcome is slow and imprecise.
Michael Brenner, Catalyst Professor of Applied Mathematics and Applied Physics at Harvard’s John A. Paulson School of Engineering and Applied Sciences, believes computation can change that. Along with graduate student Ramya Deshpande and postdoctoral researcher Francesco Mottes, Brenner and his team developed a computational framework that extracts the “rules” cells follow as they grow and interact.
The computer captures these rules as genetic networks that govern a cell’s behavior, determining how cells chemically signal each other and how physical forces cause them to stick together or separate. Brenner explained that the key was to approach cellular organization as “an optimization problem that can be solved with powerful new machine learning tools.”
Automatic Differentiation and Cell Behavior
The breakthrough came through a technique known as automatic differentiation. Originally designed for training deep learning models in artificial intelligence, automatic differentiation allows a computer to calculate the exact effect that a small change in a gene or signal has on the behavior of an entire group of cells.
As the research team describes, this method translates the incredibly complex process of cell growth into something a machine can systematically evaluate. Instead of random trial and error, the system can simulate how cells will act and then adjust parameters until a desired result emerges.
Deshpande described the potential of this approach by asking a key question: “Once you have a model that can predict what happens when you have a certain combination of cells, genes or molecules that interact, can we then invert that model and say, ‘We want these cells to come together and do this particular thing. How do we program them to do that?’”
The Long-Term Goal: Predictive Organ Design
For now, the work is a proof of concept, but its implications are far-reaching. By combining this computational framework with real-world experiments, scientists could eventually guide the development of tissues with specific functions or shapes. Mottes emphasized the long-term goal, saying, “By enabling the scaling of physics-based systems biology models, automatic differentiation offers a promising path toward achieving the predictive control needed to, in the distant future, engineer the growth of organs — the holy grail of computational bioengineering.”
He added that if models become accurate enough, scientists could one day simply request a design: “If you have a model that is predictive enough and calibrated enough on experimental data, the hope is that you can just say, for example, ‘I want a spheroid with these characteristics. How should I engineer my cells to achieve this?’”
The project has drawn support from institutions such as the Office of Naval Research and the National Science Foundation’s AI Institute of Dynamic Systems. It has also been dedicated to the memory of Alma Dal Co, a former Harvard postdoctoral researcher who contributed to early advances in this field.
AI’s Expanding Role in Medicine
This Harvard work is part of a larger trend in which artificial intelligence is reshaping medicine and biology. Lloyd B. Minor, dean of Stanford’s School of Medicine, described his first experience testing ChatGPT on a rare ear disorder he had discovered. “The way it assimilated the information and presented it was really quite remarkable,” he said. “And that’s when I realized this is not just an incremental advance. Large language models are going to fundamentally change the way we access, learn, and embrace knowledge.”
AI has already transformed medical research through tools like Google DeepMind’s AlphaFold2, which predicted the structure of nearly all known proteins. That achievement opened the door to new drugs and treatments and earned its creators a Nobel Prize. In fact, the first AI-designed drug, developed by Insilico Medicine, is now in Phase II clinical trials.
Harvard researchers have also used AI to detect cancers, predict survival rates, and improve treatment strategies. Other groups are developing AI systems that outperform traditional clinical tests in forecasting Alzheimer’s progression. These examples highlight how quickly AI is moving from experimental tools to everyday resources in labs and hospitals.
Building Models at Unprecedented Speed
One of the most important aspects of AI in science is its speed. Tasks that once consumed years of manual research can now be accomplished in days or weeks. Marinka Zitnik, an associate professor of biomedical informatics at Harvard, explained that modern models “provide instant insights at the atomic scale for some molecules that are still not accessible experimentally or that would take a tremendous amount of time and effort to generate.”
Her team’s Procyon model has already helped predict functions of proteins involved in diseases like Parkinson’s, a task once thought to be beyond computational reach. Zitnik believes these developments point to “an incredible moment that we are in,” where AI can serve as a true partner in discovery.
Toward a New Era of Computational Biology
The promise of frameworks like Brenner’s is that biology will become less a matter of trial and error and more a matter of design. Researchers will be able to state what kind of organ, tissue, or protein they want, and AI models will generate a roadmap to get there.
Challenges remain, from ensuring models are accurate to preventing data biases from skewing results. But the momentum is unmistakable. As Deshpande put it, the real test is whether models can be inverted to program cells to carry out precise tasks. That vision may still be distant, but the foundation is already being laid.
The convergence of AI and biology is creating possibilities that only a few years ago seemed like science fiction. With computation accelerating discoveries at lightning speed, medicine may be entering an age where healing begins not only in the lab or the clinic, but inside the algorithms that guide the design of life itself.
