Against Nature, On Purpose
On building what evolution wouldn't
Against Nature, On Purpose: Part I
I grew up with science and approach it with a builder’s mindset. With my PhD coming to a close, I've landed on an argument I want to make.
In the pursuit of cancer immunotherapy, and in much of therapeutic biology, we should be deliberately building non-natural systems. We must ignore evolution’s historical constraints and use biology’s own syntax (DNA, RNA, proteins) to write programs that evolution never had a reason to compile. We are not breaking the machine. We are hacking the source code.
This will sound wrong to many. Evolution is the most formidable engineer that has ever existed. Billions of years of selective pressure, an arms race with no finish line, constant refinement toward something close to optimal. The Red Queen does not rest. Every protein, every regulatory circuit, every cell fate decision a cell makes is the surviving output of that process. To a first approximation, the cell states we see are Pareto-optimal for their environment. Arguing against them looks, on its face, like arguing against the strongest engineer there is.
So why go around it?
Because our bodies weren’t built for our goals. They were built for survival and equilibrium- for staying regulated. Homeostasis is the whole organizing principle of physiology, and most of the time, that’s what keeps us alive. But the moment you want to turn the immune system loose on an established tumour, those same regulatory barriers become the obstacle. The checkpoints that stop autoimmunity are the exact mechanisms stopping anti-tumour immunity. The chromatin states locking in cell identity directly block the reprogramming we need. The feedback loops preventing runaway inflammation shut down the sustained response a cancer vaccine requires to actually work.
Evolution did build tumour-suppression mechanisms. Our immune system polices rogue cells constantly, and some species have evolved remarkable cancer resistance. All of that was optimized for "good enough": keeping the organism alive long enough to reproduce. We are aiming for a completely different threshold.
What I Learned by Trying to Reprogram Cancer
My PhD focused on reprogramming cancer cells into immune cells- turning a tumour into its own antigen-presenting cell, an in situ vaccine written in the language of the enemy itself. It produced a patent and a working demonstration that stubbornly fixed cell states could be pushed open synthetically. It also laid a foundation for other people to build on.
But the deeper lesson was about the wall.
Push a cancer cell toward becoming an immune cell and most of the transcription factors (TFs) that are supposed to do the job simply fail. They fail because the barriers in front of them (closed chromatin and silenced enhancers) are there on purpose. The cancer cell resists becoming an immune cell because its entire regulatory architecture exists to prevent exactly that transition.
Native TFs can and do work when the regulatory landscape happens to be favourable. I do not discount them. But they fail at the scale of universality. Against the heterogeneity of a solid tumour, native TFs at their natural expression levels just don't have the horsepower to force the transition consistently.
This led me to engineer synthetic TFs using induced proximity- physically forcing factors together that would never naturally sit in the same regulatory space and with stronger activation domains. Their combined activity clears the potency bar that chromatin barriers set. Nature keeps some factors apart because homeostasis depends on it. If the therapeutic goal needs their combined force at a specific locus, at levels nature would never tolerate, you build that proximity by hand.
This is the microcosm of my entire argument: deliberately breaking the natural order at the molecular level because the goal demands a configuration evolution never selected for.
Conventional wisdom says respect the regulatory architecture by finding the right native cocktail and working with the given system. Working within the system caps you well short of a universal cure. You get there by building what nature wouldn't, deploying it where nature wouldn't, at a potency nature would never allow. You install functional modules that bypass the entrenched fate directly, opening chromatin evolution shut, at expression levels evolution selected against, gated by disease-responsive elements so the whole thing only fires where it's needed.
Am I working against biology? Not exactly. I am still using biology's syntax, just deploying it somewhere evolution never intended. That is deliberately hacking the system inside a constrained context toward a defined goal.
How the Execution Changes
My PhD was conducted in a heavily wet-lab environment, and the rigor it demanded was essential. The bench is still where intuition gets built, and it remains the final arbiter of what is actually true. I'm not walking away from it.
But while one foot was executing science day by day, the other was exploring outside the bubble. I spent the back half of my PhD watching the world expand. Computational biology transitioning from an accessory to a foundation. I went looking for that because I could see the ceiling of what I already knew how to do, and I wanted to know what was above it.
If I were to start again, I would not begin at the bench. I would begin with a Socratic accounting outside the lab entirely: What are we actually trying to achieve? What are the deficiencies in the traditional solutions? What non-natural systems must I build to bypass them? Only after that would I move to computation to simulate and filter the highest-probability candidates. Only then would I hit the bench, at which point the hypothesis space is already small and the work is focused.
Biological engineering has spent decades as a brute-force discipline: build something, test it, watch it fail, guess at what went wrong, iterate. What’s changing isn’t the wet lab- it’s that reasoning and execution can now run in parallel. We don’t stop building physically. We stop building blindly.
Where the Domain Expert Stands
I am a biologist. My hands are still my primary instrument. But I've spent enough time working computational tools into that process to know where they break down and how to frame a question so the tool actually helps.
The real advantage is knowing which model to point at which question, and being able to tell whether the answer it gives back makes any biological sense.
Molecular design is going to get commoditized. Within a year or two, generating a tighter binder or a more potent TF will be routine. Once an algorithm can produce a better X in seconds, the only thing left of actual value is deciding what Y to build instead, and knowing why Y matters.
Foundation models are powerful hypothesis generators. But a hypothesis is not an intervention. The gap between a computationally generated prediction and something that functions inside a living system is enormous. Closing that gap belongs to people who've spent years at the bench learning exactly why things fail. The people standing at the seam between the digital model and physical reality are deciding what's actually going to work. That's where the leverage sits right now.
We’re in a strange, narrow window: the computational tools are finally strong enough to carry the weight of real hypotheses, but they still depend entirely on a biologist’s physical intuition to become real.
What’s Next
The foundation of many diseases is mutational: errors in the code that produce errors in function. Correction and targeting are the fundamental interventions. Programmable biology at close to single-nucleotide resolution isn't a research niche anymore; it's becoming the infrastructure medicine runs on. Nearly every therapeutic modality I care about bottlenecks on hitting the right genes and delivering the right payload to the right cell at the right time.
But prediction without manipulation is merely academic. At some point, we need a product, a tangible object that enters a cell and changes its behaviour. The models tell us where to aim. The product is what hits the target.
The blood compartment is where I started before graduate school, and it is the same problem in a different organ. The ability to reshape and clear the blood compartment without destroying the patient first has long been a holy grail of hematology. The blood system is homeostasis incarnate, a tightly regulated equilibrium that keeps you alive, but also fiercely resists the targeted reconstitution that would cure sickle cell, thalassemia, or bone marrow failure. Its natural stability is the barrier. Rebuilding it means writing conditioning regimens that clear selectively, engineer stem cells that engraft on your terms, correct genes precisely, and manage the resulting immune fallout.
This logic applies far beyond blood. A person gets sick. Something is wrong at the level of their cells, their genes, their immune system. And then what?
Increasingly, the answer is going to be programmable biological interventions, designed from first principles, deliberately non-natural, tightly controlled, and doing things the body would never do left to its own devices.
So my next phase is creating inside a loop that keeps running: the bench teaches the model, the model reshapes the next experiment, the experiment reveals more biology, and the biology redefines the question. I want to build the tangible things that come out the far end of that loop- engineered TFs, gene editors, targeted delivery systems, and programmable circuits. None of this philosophy means anything if it doesn't eventually become a product that cures something.
Evolution built a remarkable optimum. I respect it, and I have spent years learning its language. But I am not finishing this PhD to refine what nature has already done. I am finishing it to write what nature never would.
Against Nature, On Purpose: Part II - Let’s Pick a Disease
Cancer is the right place to start. I have spent years researching it, from the developmental origins of malignancy to the engineering of synthetic immune responses against it. When I actually think through what it would take to cure it, I keep landing on the same first principles.
Cancer is not an event. It is an accumulation. Mutations build up over a lifetime from replication errors, environmental insults, metabolic byproducts, and the slow erosion of genomic fidelity we call aging. Repair machinery exists, but it has gaps. Some regions are more exposed. High-turnover tissues accumulate damage faster than others.
Most of these mutations are passengers. They sit in the genome and do nothing consequential. But some land in driver genes: oncogenes, tumour suppressors, regulators of proliferation and apoptosis. When enough drivers accumulate in the right combination, the cell crosses a threshold. Its integrity is compromised. It begins to behave in ways the body’s regulatory systems cannot contain.
This is where evolution's design shows its limits. p53 and immune surveillance catch most rogue cells before they become a clinical problem, but they were built for "good enough," not for zero cancer over an eighty-year lifespan.
That produces a spectrum. At one end: clonal hematopoiesis of indeterminate potential (CHIP), where blood stem cells acquire mutations and expand clonally without ever crossing into cancer. At the other: full malignancy, a cell that's evaded immune surveillance, resisted apoptosis, hijacked its microenvironment, and started proliferating without limit. Between CHIP and leukemia sits a continuum defined by how many of evolution's barriers have already been breached.
The question is not simply “how do we kill cancer cells.” The real question is where on that continuum to intervene, with how much force and precision, to stop the transition to malignancy. The answer depends heavily on where the cancer actually lives.
Blood: Replace It
Blood is ubiquitous. It traverses and bathes every organ, carries the immune system to every tissue, and connects every compartment of the body through a single circulating network. I trained in blood development and stem cell transplantation before graduate school, and it left me with one fixed idea: blood isn't just a tissue. It's the body's operating system for defense and systemic communication.
When the disease is the blood itself, the logic is almost clean: replace the blood.
I do not say this lightly. Transplantation is brutal. Conditioning regimens that ablate the existing marrow carry significant mortality. Donor matching is imperfect. Graft-versus-host disease is real. The entire process is a controlled demolition followed by a rebuild, and it can kill the patient in the process of trying to save them.
The central problem in curing blood disease is not conceptual. It is logistical and biological: can you clear the diseased blood selectively enough, and replenish the healthy blood reliably enough, without destroying the patient in between?
This is where the engineering must go. In vivo conditioning that selectively targets diseased clones without ablating everything. Blood expansion protocols that generate enough functional cells for transplant without depending on perfect donor matches. Gene correction in hematopoietic stem cells with enough efficiency and precision to fix the mutation at the source. Every one of these is a non-natural intervention (writing programs the body was never designed to run), and every one of them is a solvable engineering problem if pursued with sufficient focus.
Solid Tumours: You Cannot Replace a Brain
But you cannot replace a brain. You cannot reconstitute a pancreas from stem cells the way you can rebuild a blood system. Solid organs do not regenerate on command.
When the cancer is solid, embedded in a tissue you cannot simply remove and replace, the strategy must shift. And it shifts, ultimately, back to the blood. Not to blood replacement, but to what the blood carries: the immune system.
The strongest evidence for durable cancer cures we have comes from immunotherapy: checkpoint inhibitors releasing the brakes and CAR-T engineering immune cells against specific antigens. These converge on a sustained, memory-forming immune response that clears the cancer and stays primed in case a surviving cell resurfaces. The underlying requirement is identical across every modality- turn the immune system on, point it at the right target, and keep it running long enough to actually finish the job.
The Trap of Incrementalism
What is five extra months from a transient drug that prolongs survival incrementally? What is the value of the tenth derivative of the same target, the next-generation inhibitor competing in an already crowded space for marginal improvement?
Five months is five months. Five days is five days. Time means something to anyone facing mortality, and I do not want to minimize that. If we're asking where a whole field should point its collective effort, the answer isn't more incremental extensions.
Evolution will beat us if we try to compete incrementally. This is not a metaphor. It is the literal biology of resistance. Tumours evolve under selective pressure. Every targeted therapy creates a bottleneck, and the surviving cells, the ones with the resistance mutation or the phenotypic plasticity to evade, expand to fill it. We introduce a drug; the tumour adapts. We introduce a second-line therapy; it adapts again. We are running the Red Queen’s race against a system that evolves faster than we can iterate.
The alternative is not to compete. It is to monopolize. To apply Peter Thiel’s framework to biology: do not build a marginally better version of what exists. Build something fundamentally new, zero to one, that changes the terms of the contest entirely. Blood replacement that eliminates the diseased compartment. Systemic immunity that provides durable memory. These are not incremental improvements. They are category shifts grounded in first principles that cover a larger surface area than any single-target inhibitor ever could.
The Geography of the Cure
The strategy must also match the patient’s reality. An ex vivo cell therapy- pulling a patient’s cells out, engineering them, putting them back in- is a boutique product. It requires specialized centers, cold chain logistics, and enormous cost. It works for those with access, but it does not scale to the billions of people who will get cancer in countries without that infrastructure.
For most people, the therapy has to work in vivo- something injected that activates immunity, hunts down the tumour, establishes memory, and produces durable protection without a specialized facility behind it. The underlying biology doesn’t change; the delivery has to. Designing for that kind of access from day one is itself a first-principles decision that shapes everything downstream.
Blood replacement where blood is the disease. Systemic immunity where the tumour is embedded. Modalities built for the context they actually have to survive in. Every one of them is a non-natural intervention. Who gets access to them is a design constraint from the start.
Against Nature, On Purpose: Part III - The Build
The philosophy is set. The target is chosen. What’s left is the product.
There is a trap in synthetic biology that is easy to fall into: overengineering. We learn to write genetic circuits and suddenly we want to build biological Rube Goldberg machines: systems with nested logic gates, cascading dependencies, and elegant architectures that look beautiful on a whiteboard but fail catastrophically in vivo.
True biological design is ruthlessly reductive. It is the minimum viable deviation from nature required to achieve a non-natural result. Take nature's logic, strip out its limitations, and keep only what actually works.
Universality Over Personalization
The consensus in a lot of modern medicine is that the future is personalized oncology and treatments built entirely around an individual's genome. There's real science behind that. Tumour heterogeneity and patient-specific neoantigens are real.
But I want to push back on whether personalization should be the central organizing principle of how we build.
If you can build for universality, why personalize?
In a lot of cases, personalization is an admission that the underlying tool is lacking the potency and precision to work across the full range of human variation. We sequence the tumour, find its specific mutations, and build a therapy that works for that one patient against that specific tumour. The next patient has different mutations. The next tumour has different antigens. Every patient becomes a bespoke engineering problem, and the cost, the timeline, and the scalability constraints of that approach are staggering.
What if the question were different: what stays constant? What functional requirements don't actually change from patient to patient, or tumour to tumour?
If the goal is sustained systemic immunity, Signal 1 (antigen presentation), Signal 2 (costimulation), and Signal 3 (cytokine polarization) don’t change from person to person. The chromatin barriers in a solid tumour don't differ in kind, only in degree. The need for potent TF activity at locked loci is universal. The need for targeted delivery to diseased tissue is universal. Build for those invariants and you get something that works broadly by design. You are forced to solve the hardest barriers directly instead of tailoring around them, but the payoff is a platform instead of a one-off.
The Architecture, Step by Step
What does a universal product actually look like, broken down to first principles?
A patient takes a pill or gets an injection. It enters the bloodstream, reaches the target tissue, crosses the cell membrane, releases its cargo inside the cell, and that cargo makes it to the nucleus. Once there, it turns genes on or off, integrates a sequence, or repairs a mutation- and then it degrades. The patient goes home.
Each step is its own engineering problem, solvable on its own terms: membrane penetrance, endosomal escape, nuclear localization, transcriptional control at a specific locus, stable integration where permanence is needed, and controlled degradation once the job is done. Each is a node that can be optimized independently and then assembled into a working system.
The key is that this sequence is modular. The same delivery chassis that gets a synthetic TF into a cancer cell can get a gene editor into a hematopoietic stem cell. The same nuclear targeting logic that directs cargo to a specific locus in a tumour can direct cargo to a specific locus in a degenerating neuron. The architecture is universal even when the payload changes. First principles at every node. Universality across applications.
Understanding Through Building
There is a way of doing science that is undervalued: learning by building the opposite of what nature intended.
I didn't set out to map the regulatory landscape of immune gene programs in my own work on constitutive TFs. Forcing TFs to operate at levels and configurations nature never permits showed me where the brakes actually are. Points of resistance aren't just obstacles. They tell you what the cell considers important enough to defend, mapping regulatory logic that passive observation would never surface.
This is a general principle. Build toward your goal aggressively, and the biology will show you where it resists. Those points of resistance are not just obstacles. They are data. They tell you what the system considers important enough to protect. They map the regulatory logic that no amount of passive observation would reveal as clearly.
You don’t need to fully understand a system before you build against it. Building against it is often how you come to understand it.
The same logic should apply to how we chase universality as a research question. We need screens for what works across the widest range of contexts: which TFs open chromatin broadly. We need to identify delivery methods that achieve penetrance across tissue types, and map regulatory elements that give disease-specific activation regardless of their tissue of origin. These take large-scale, first-principles experimentation to answer.
Beyond the Familiar
The same first principles- penetrance, cargo delivery, targeted gene manipulation, permanence- apply well outside oncology.
Reproductive biology has already shown functional sex cells generated from somatic cells, with massive implications for genetic disease prevention. Aging research is exploring reprogramming to restore youth to existing cells. Neurodegeneration may have the hardest delivery problem in all of medicine, because the blood-brain barrier is about as tightly regulated a gate as evolution has ever built.
This is where interdisciplinary convergence becomes strictly necessary. The next breakthrough in delivery might come from a marine biologist studying how certain organisms penetrate cell membranes. The next TF architecture might borrow logic from a signalling cascade in a plant. We do not know where the foundational discoveries will come from, but the history of biology tells us they rarely come from the field that needed them most. With AI and large-scale screening now able to search across biological domains faster than any person could, we're in a good position to find those answers.
The Product Is the Point
I am not interested in architectures that exist only on paper. The argument means nothing without the build.
Permanence and universality are the design principles that matter to me: systems that work broadly and permanently. We want to extract nature’s logic while discarding its limitations. The smallest deviation from natural biology that still gets maximum potency, engineered node by node from membrane to nucleus.
We have the syntax and the tools. The computational reach to search a massive design space that was previously inaccessible. What's left is the nerve to build it, and the discipline to build only what actually works.