Risk-rewards in pharma will change with Cell Cinema, and here's how.

I keep coming back to one question: what becomes worth attempting when biology becomes easier to read? Most conversations about AI and drug discovery are still stuck on generation.

• Can AI design a molecule?
• Can it suggest a target?
• Can it optimize a peptide, protein, antibody, or small molecule faster than a human team would?

The answer is increasingly yes. But the deeper shift happens when the biological feedback loop becomes fast and information-rich enough that those generated ideas can be corrected by reality.

In software, the feedback loop is tight.

1. write code
2. run it
3. see what breaks
4. fix it
5. run it again

This is one reason software compounds so quickly.

In biology, the feedback loop has been slow, expensive, destructive, and often indirect. It involves: perturbing a system, waiting, stop experiment, stain it, sequence it, image it, or run an endpoint assay, and then reconstruct what you think happened. Sometimes that works. Sometimes the most important part of the biology happened in between the timepoints, in a cell state you did not measure, through a compensatory pathway you did not expect, or in a form of toxicity that only becomes obvious when the system is watched over time.

Cell Cinema is what changes when that feedback loop starts to look more like a movie and less like an autopsy. I do not mean that we are simply making prettier images of cells. The real point is that living cells move through states, and many of the most important drug effects are not static.

In today's world:

• A compound can look acceptable at one endpoint and still create a dangerous transient stress response earlier.
• A molecule can hit its intended target and still push mitochondria, morphology, metabolism, membrane behavior, or signaling dynamics in the wrong direction.
• A therapy can appear to "work" in a narrow assay while the broader cellular system is telling a more complicated story.

Tomorrow:

If you only ask one question at one timepoint, you get one kind of answer. If you watch many dimensions of the same living system over time, you get Cell Cinema. This is why I think the serious discussion after massively scaled, multiplexed Cell Cinema is not "what can we see?" Rather "what risks become small enough that humanity becomes willing to try harder things?"

Drug discovery has always had a strange risk-reward shape.

The reward for being right is enormous. A successful drug can change the standard of care for a disease, create a generational company, and alter the lives of millions of people. But the risk of being wrong is also enormous, because you often find out late. You can spend years building conviction around a target, molecule, indication, and mechanism, only to discover in humans that the biology did not translate, the drug was toxic, the effect was too weak, or the disease was not being modulated in the way your model implied. This has made the whole industry more conservative than people admit. Pharma is not conservative because scientists lack imagination. The evidence has just historically been far too weak to justify enough imagination. When the cost of being wrong is catastrophic, rational people cluster around validated biology, known modalities, precedent, and mechanisms that have already survived some contact with the clinic. The industry says it wants novelty, but structurally it punishes novelty whenever the measurement layer cannot make that novelty legible. Only one thing can change this: readout.

Derisking the battle of finding a cure.

The first wave is simple derisking. This is the least glamorous part and probably the most commercially important place to begin. If a drug has toxicity, off-target effects, or an ugly cellular phenotype, I want us to find that earlier. About 60-70% of a 90% failure rate in drug discovery are clustered around tox and off-target effects.

But toxicity or off target-effects are not singular things, they were measured as singular things. Tox can be: mitochondrial dysfunction, stress pathway activation, and so on. For targets, it can mean the molecule is acting on a target we did not intend, but it can also mean that the intended target is wired into biology in a way that creates second-order effects we did not predict. These are measured as singular things.

Cell Cinema sees them as (1) live, and (2) high-dimensional vectors. This matters. Cells do not simply jump from healthy to dead, instead they often are benefitted by evolution, and compensate, adapt, fail, recover, or enter intermediate states. A destructive endpoint assay can miss those trajectories because it collapses the process into a final measurement. A living readout can show the path where a lot of the truth is; in biology, this truth is simply too high-dimensional for anything other than a GPU.

In the beginning, derisking may look like a very practical customer problem. Then the problem starts scaling. Intelligence, too cheap to meter. That scale creates the highest impact: behavioral. If a company can discover a bad liability earlier, the downside of trying a program changes. The failure still hurts, but it hurts at the right time. Moving truth earlier in the development timeline is one of the most powerful things you can do in drug discovery, because late truth destroys companies and early truth compounds. Once THAT starts happening, people become willing to take better risks.

Giving eyes to the blind.

That is the second wave: novelty becomes more financeable. Novel mechanisms have always had a penalty attached to them. If there is no precedent, there is no easy prior. If there is no easy prior, every committee, investor, pharma BD team, and internal champion has to decide whether the evidence is strong enough to overcome the discomfort of the unknown.

A lot of good biology dies in that gap, forming a systemic valley of death. Not because the idea was wrong, but because the system could not tell whether it was wrong early enough. Cell Cinema helps separate "novel and wrong" from "novel and early." And that distinction is everything, because we live in a strange world where novelty was punished instead of rewarded, DUE TO risk. A new mechanism should never be punished merely because the old measurement stack cannot evaluate it well. The measurement stack needs to be in the loop.

If we can watch the relevant human cells respond over time, across enough biological dimensions, then novelty becomes something we can reason about instead of fear. We can see whether the perturbation moves the cell toward the desired state, whether the response is durable, whether the system compensates, whether stress appears, whether related cell types behave differently, and whether the phenotype resembles something we would actually want in a human. This is the point where the economics start to change in a very, very non-obvious way. Better readout does not just reduce bad bets. It increases the expected value of bold bets. If the probability of success rises while the reward remains high, then entire categories of programs that used to look irrational begin to look rational.

A cliff looms, NOW.

This is NOT a long-term hypothesis. It is especially important right now because the rest of the industry is already pushing more candidates into the funnel:

1. AI drug discovery is producing more hypotheses and deals with pharma moved from 0 to 105 in one year as it started.
2. GLP-1s and the peptide boom have reminded everyone that modalities can move from niche to massive once manufacturing, demand, and clinical validation line up.
3. NAMs are becoming more serious as regulators open the door to human-relevant non-animal methods.
4. The virtual cell is becoming less like a metaphor and more like an organized ambition.
5. The patent cliff is forcing large pharma to find new sources of growth instead of living forever on yesterday's franchises.
6. All of those forces create pressure on the same bottleneck: proof.

More design without more proof just creates a larger pile of untrusted candidates. The scarce thing is shifting from "can you generate something?" to "can you tell whether the generated thing matters?" That is why I think readout becomes one of the most important layers in the AI age of medicine.

Readout brings in the AGI of biology.

The AI version of this is not that a model wakes up one day and cures disease by thinking harder. The real version is that intelligence becomes coupled to experimentation tightly enough that the system learns from biology at a speed and resolution we have never had before.

1. AI proposes perturbations.
2. Living human cells respond.
3. Cell Cinema captures the response as a dynamic, high-dimensional movie.
4. Models learn not just from labels, but from trajectories.
5. The next experiment is better because the last experiment was richer.

Over time, the system becomes better at asking biology questions. This is the third wave, and it is the biggest one. At that point, drug discovery becomes less like asking "does this molecule bind this target?" and more like asking "what state is this cell in, what state should it be in, and what intervention moves it there without creating unacceptable damage elsewhere?" That is a fundamentally different search problem.

Cells are not just bags of targets. A beautiful model of the cell without dense experimental correction is still mostly an elegant guess, a 'virtual cell.' The combination of model and readout is what matters. It brings: an increasingly predictive model of cellular response connected to a scalable system for observing actual living biology is a software-like revolution.

This is why I do not think of Cell Cinema as only an assay business. The much larger prize is building an evidence layer for AI-native biology. The world is going to generate more candidate drugs, more modalities, more edits, more combinations, and more mechanistic hypotheses than it can evaluate using the old stack. Someone just had to build the system that tells the world which of those ideas are real. The reward will shift toward whoever owns that evidence. In the previous era, much of the value sat with companies that had clinical development capacity, regulatory expertise, capital, manufacturing, and commercial reach. But as generation becomes cheaper, validation becomes more valuable. If molecule design becomes more commoditized, then the hard-to-recreate asset is the trusted biological readout.

A semiconductor-like moment for Cell Cinema will change not just Precigenetics. It brings hope.

Precigenetics starts with the practical wedge because that is how important companies are built. You do not begin by declaring that you are changing the future of medicine. Toxicity and off-target derisking are urgent because late failure is absolutely brutal to both economics and, frankly, hope. If we can help customers understand earlier which drugs are dangerous, that is already valuable.

But the positive externality is much larger than any of this, and it compounds.

• Every weak program killed earlier frees capital, people, and time for something better.
• Every strong program that gets better evidence becomes easier to fund.
• Every novel mechanism that becomes more legible makes the whole industry slightly braver.
• Every human-relevant dataset that improves a model makes the next experiment more informed.
• Every regulatory pathway that accepts better non-animal evidence reduces the time and friction between hypothesis and human relevance.

Importantly, this HAS TO compound through society. Suddenly, humanity is more willing to try. More diseases become economically addressable, the valley of death improves and the frontier of medicine moves because the frontier of evidence moved first.

This is also why the risk-reward ratio matters more than cost alone. People often talk about lowering the cost of drug discovery, and that is obviously important. Risk is, however, what has been secretly governing behavior. Therefore, the real unlock of lowering uncertainty lies in the places where uncertainty changes decisions.

In the medium term, I think risk goes down before reward goes down. That is a very unusual window. If drug discovery became easy overnight, then eventually the rewards would compress. Successful drugs would become less scarce, some categories would commoditize, and the economics of therapeutics would look different. That is not where we are. We are nowhere near saturation. Most diseases are not solved, and biology is still barely understood. The world still has far more unmet need than credible therapeutic programs.

So if Cell Cinema, Cleopatra, and related downstream technologies lower the risk of making drugs while the rewards remain high, the immediate effect is not commoditization. The immediate effect is MORE, BETTER, AND MORE AMBITIOUS ATTEMPTS.

That is the world I am trying to build toward. A world where AI can generate hypotheses, but biology gets to answer quickly. A world where novel mechanisms are not dismissed because our tools are too weak to understand them. A world where the cost of being wrong is lower, the value of being right is still enormous, and the system becomes brave for rational reasons. I prefer this world very much, because it benefits those who are far from this industry.

The most important technologies do not only make existing work faster. They change what people believe is worth doing. Cell Cinema changes that belief because it changes the evidence underneath it.

— Parmita Mishra | Founder & CEO, Precigenetics