The golden age of many fields of science — from astronomy to economics — arose when the empirical observations converged with the analytical ability to make accurate predictions. Today, immunology has reached that crossroads: The conjunction of AI models plus the availability of computing power is leading to a shift from watching the immune system work to actually predicting what it will do next.
The immune system is one of the oldest components of multicellular organisms, and it evolved to keep our bodies parasite-free and healthy through reproductive age — which 200,000 years ago was likely between 15 and 20 years old. With longer lifespans today, humans encounter illnesses that the immune system did not evolve to handle, including rheumatoid arthritis, multiple sclerosis, most cancers, Alzheimer’s disease, and, of course, the obvious one, aging.
But with today’s tools, we can harness the immune system to prevent or even cure those illnesses. The convergence of AI and machine learning with single-cell biology and the iPSC (induced pluripotent stem cell) revolution has opened a new era of engineering the immune system. Biohub’s team of AI researchers, immunologists, and bioengineers aims to rewire immune cells to detect disease, report on physical changes, and deliver therapeutics.
Here are the six strategies we’re pursuing to reprogram the immune system:
1. Harnessing the immune system’s natural abilities
In April 2025, a prestigious Breakthrough Prize, founded by Priscilla Chan, Mark Zuckerberg, and others, was awarded to two researchers for their discovery that immune cells infected with the Epstein-Barr virus play a major, causative role in multiple sclerosis, a chronic disease of the central nervous system that affects 2.9 million people worldwide and has no cure.
It is the latest of many discoveries that clearly indicate that too much or too little inflammation plays an important role in major diseases that affect most of the population. At Biohub, we think we can reverse many of those negative immune effects by using immune cells’ innate, natural abilities: Immune cells can go anywhere in our bodies, have amazing sensor mechanisms, and they can trigger or block inflammation.
We can not only train those abilities to fight or prevent disease, but to maintain health throughout life. As we age, our bodies become less effective at certain functions, largely because the immune system no longer scavenges and eliminates senescent cells — cells that have aged and no longer divide, but do not die when they should. Recognizing senescent cells and eliminating them to make space for new growth and less inflammation could be achieved by modulating the immune system.
2. Learning the immune system’s internal logic with AI
Cancer evolves over long periods of time by slowly accruing mutations that don’t trigger an immune response. Then, tumors recruit parts of the immune system that block the effects of other immune cells that would attack the cancer. We aspire to teach immune cells to revert that effect — but first, we need to understand the logic by which cancer recruits and/or reprograms immune cells to create an immunosuppressive barrier.
One way we’re doing that, for instance, is to generate 50 million single cells from macrophages and fibroblasts — the latter are not immune cells but key immunomodulators — and systematically silencing each of the proteins with regulatory or signaling activity using CRISPR technology. We can then observe and document how each genetic change affects a cell’s function, and with that data, we will build a predictive AI model to tell us exactly how to reprogram those cell types to an anti-tumor state. Then, instead of guessing how to reprogram cells to fight tumors, we’ll have an AI-powered “instruction manual” that says, for instance, “Silence these three genes and activate these two genes to transform a macrophage from a tumor helper into a tumor fighter.”
Similarly, if we want immune cells that can go to a specific organ on demand — a process known as organ tropism — we need to understand the rules that determine that outcome, so we can engineer them to go where we want them to go, such as to the pancreas, brain, or liver, and report back on what they have seen.
3. Reprogram immune cell states at scale
With 20,000 genes encoding a million different protein isoforms, the number of potential interactions between proteins, DNA and RNA that may drive the behavior of human cells is simply staggering — larger than the number of atoms in the universe. By restricting the search space, using what we already know about biology, we can reduce the number of actual interactions to only a couple million, thus making the use of transformer models feasible. This is how other successful AIs have worked, like AlphaFold, which uses priors — based, for instance, on evolutionary conservation and other protein properties — to predict their folding.
As we learn and apply AI to prior knowledge of the immune system, we can use it to rewire immune cells at scale. For example, normally, regulatory T cells help prevent inflammation and autoimmunity in tissues exposed to external organisms, like in our gut. Tumors, however, learn to recruit regulatory T cells to block the immune system from detecting them. In work published in Cancer Cell, we were able to reprogram regulatory T cells so that they could no longer be recruited by a tumor. Using machine learning approaches, we predicted that silencing a single gene, Trps1, would reprogram regulatory cells produced by the bone marrow so they can no longer be recruited by the tumor. In our experiment, preventing this recruitment to tumors induced spontaneous remission. Next, we predicted which drugs would recapitulate that effect, and observed that minimal amounts of a predicted drug could inhibit tumor growth in vivo in combination with immunotherapy, while both drugs were completely ineffective as monotherapies. That’s just one example of the ways we are identifying the triggers to reprogram immune cells at scale in the human body.
4. Record disease signals in immune cell DNA
If we want to move from fighting disease to preserving health, we need early signs of distress identified immediately. One way to do that is by instrumenting immune cells so, when they see a particular signal of distress, they write it into their DNA like a memory device. Then those cells lyse, or break apart, and that DNA gets shed into the blood. With a simple blood draw, one can then read what is actually happening in each of our organs.
Harris Wang at Columbia University Medical Center, one of Biohub’s funded investigators, has already shown how one can write binary code into DNA, as a first step to building molecular memory systems. This approach could eventually turn whatever the reprogrammed cells are sensing into a pathology report on the state of health or disease of an organ.
5. Move beyond mice to human systems
A lot of immunology work is done in mice, but the mouse immune system is very different from the human immune system. Biohub investigator Gordana Vunjak-Novakovic is creating an “organs on a chip” device that addresses this issue. This includes an iPSC-derived human bone marrow that can generate myeloid cells and is now being refined to also produce lymphoid cells. These bioreactors also grow organs — a liver, skin, a pancreas — all connected by blood flow, so we can study exactly what makes immune cells traffic to one organ versus another. These technologies are simply remarkable and a game-changer in modeling cell behavior in a human-relevant context.
Biohub investigator John Tsang at Yale has launched the Human Immunome Project, an amazing initiative to do long-term longitudinal profiling of immune responses to challenges like vaccination or infection in populations around the world. From that project, the scientific community will gain single-cell data, genomic data, and physiologic and clinical readouts from thousands of people. That human immune data will be a landmark change in systems human immunology and how we can explore it with AI.
6. Build communities across disciplines
Bioengineering the immune system is not something that can be done in isolation. It requires a large community across Biohub and with external parties. For instance, we don’t have clinical work being done at Biohub, so we need strong ties to communities that do clinical work, in academia and industry.
During my entire career, I’ve striven to gather communities together to solve problems bigger than any lab can solve individually. Today, we are creating a community of synthetic biologists, systems immunologists, and experimental biologists who use AI and machine learning to predict what immune system cells will do. We launched a program in synthetic biology — perhaps the largest synthetic biology community focused on immunology in the world. Seven leading labs are already funded, and we have completed an RFA call, with the grantees to be announced.
Creating these communities is our top priority because it’s the only way we’ll tackle this complicated, ambitious goal.
Andrea Califano serves as president of immune cell reprogramming and head of Biohub, New York, where he is focused on reprogramming the immune system to maintain health throughout human life. His work spans systems biology, computational approaches to predict cellular behavior, and the integration of synthetic biology with immunology research.