Tal Zaks - Bridging Science, Medicine, and Returns - [Invest Like the Best, EP.406]
Summary
Biotech’s opportunity is expanding, but its returns remain governed by two stubborn unknowns: whether the biology is relevant to disease and whether a drug can alter it safely in the right tissue. Tal Zaks reduces ROI to capital required, time to proof, and probability of arrival. AI is accelerating drug discovery, but without a holistic human model, “there’s no shortcut” past clinical testing.
Biotech VC cannot copy software’s power-law tolerance for failure because its upside is smaller. Where tech might survive on one 1,000-to-1 outcome in 100 bets, biotech targets 3-4x, regards 5x as great, and rarely exceeds 10x. Zaks therefore needs incremental de-risking: he walked from a vaccine whose phase 1 would not answer the key question because only phase 3 could.
Moderna’s decisive asset was not one drug but a platform with enough capital and strategic freedom to discover its best applications. Before COVID-19, its mRNA technology had generated neutralizing antibodies in humans against eight viruses in eight attempts; COVID became the ninth. The same platform now underpins a personalized skin-cancer vaccine that cut recurrence by about 50% in a randomized phase 2, with phase 3 enrolling.
COVID vaccine speed came from accumulated readiness, compressed coordination, at-risk manufacturing, and smarter trial placement—not from eliminating evidence. Government discussions that usually took weeks or months happened within a week, private capital funded commercial expansion before government purchase commitments backstopped it, and predicted hotspots produced in three months the events expected to take up to a year. “We were very well prepared as a company with a platform.”
Zaks is categorical about the mRNA vaccines’ benefit and safety evidence, but concedes that scientists overstepped the boundary between scientific advice and democratic legitimacy. He calls them “the most well-studied medical intervention in the history of mankind,” while acknowledging myocarditis and other rare findings were detected and labeled. His postmortem: science supplies evidence, not an ethical mandate—“forcing people to take vaccines” required public consent that scientists could not dictate.
Nucleic-acid medicines make drugs increasingly information-like, lowering the marginal cost of each new program once delivery works. Conventional mRNA is transient, while mRNA can also be used as an intermediary to make a lasting change to DNA, potentially replacing injections every two weeks with one to three treatments. That promise brings legitimate questions about autonomy, lifelong changes, and effects on future generations.
The bull case for AI in healthcare depends less on digital-human simulation than on institutions adopting integrated tools with aligned economics. Zaks’s bear case is scattered products for narrow workflows; his bull case reverses the productivity damage inflicted by electronic records. Digital twins remain “a little bit the realm of science fiction,” while today’s warning is that even FDA-approved computational pathology is used on only a tiny fraction of tissue-microsection slides because nobody has solved who funds deployment and captures the return.
Deep dive
1. Medical innovation survives only when capital comes back
Zaks’s State of the Union is optimistic on science but guarded on translation, public trust, and willingness to pay. Better biology does not remove the unpredictability of medicines, and the current political environment can weaken the commercial engine needed to fund innovation.
His correction to a National Academies panel was deliberately blunt: “Reimbursement is what I do when I put in my expenses.” Investors support the mission, but venture capital’s actual job is equally plain: “People give us money, and a number of years later, we need to give them more money back.”
The social bargain extends beyond the patent window. Zaks credits both NIH-funded research and America’s commercial infrastructure, then points to Teva and his father’s post-heart-attack medicines: after branded pricing ends, generics can deliver decades of global benefit “at pennies.”
2. Biology refuses engineering-style certainty
Three variables determine return: how much capital is required, how long the journey takes, and the probability of reaching the destination. Biotech’s probability is especially opaque because investors must separately underwrite biology—does the target actually relate to the disease?—and pharmacology—can the drug reach and alter it safely?
Moderna taught Zaks that engineers emphasize vision and process. An engineer asked about reaching the moon can map what it will take; a physician asked about curing cancer must answer, “I don’t know. I mean, I can think of some approaches, but we gotta go try them.”
That uncertainty shaped Moderna’s personalized cancer vaccine. Patrick initially imagined preventing cancers before diagnosis; Zaks corrected him: the vaccine is made after an early skin cancer is removed, aiming to prevent recurrence. In randomized phase 2, it reduced recurrence by about 50%; phase 3 was enrolling.
3. AI improves discovery before it eliminates clinical risk
Zaks divides translation into three stages: validate the biological target, discover an entity that changes it, and prove benefit in people. Target biology is advancing through both large datasets and “old school grunt work,” but relevance to actual disease remains the first hurdle.
Drug discovery is receiving the clearest AI boost. AlphaFold-style structure prediction is becoming commoditized; related tools can propose new chemistries, protein or nucleic-acid entities, and even lipid nanoparticles designed to deliver mRNA into different tissues.
Clinical development remains the bottleneck. Better datasets can predict control-arm outcomes and make trials smaller or shorter, but “we don’t have a holistic model of a human being.” The gold standard remains giving the intervention to people and showing that their outcomes beat those of people who did not receive it.
4. Venture underwriting requires visible, financeable proof points
Zaks begins with the scientific thesis, capital required, next value inflection, probability of reaching it, and the team capable of doing so. He also asks whether his operating experience adds something specific; he does not want to invest where he can contribute neither expertise nor informed judgment.
One vaccine opportunity failed that test despite interesting science. Its phase 1 would probably not de-risk the thesis, meaning investors would need to finance it through phase 3 before learning whether it worked—“a quantum of capital and a time horizon” disproportionate to the risk.
Initial diligence takes roughly four, six, or eight weeks; ownership can last years. After investing, the diligence team recedes while the partner works with management and the board, making reserves, follow-on financing, and a syndicate with complementary skills central to the original decision.
Zaks separates three pools of money: government for the greater good, philanthropy, and return-seeking capital. Some projects properly belong in the first two because their scale, duration, or risk cannot produce the return owed to venture LPs.
5. Multidisciplinary execution starts with breaking language barriers
A drug program can run for ten years without anyone from the original project team remaining at approval. Pharma’s institutional achievement is coordinating biologists, physicians, chemists, engineers, manufacturers, financiers, and others so the work survives those handoffs.
Drawing on John Ralston Saul’s Voltaire’s Bastards, Zaks argues that expert silos protect themselves through language. Science and reason are not inherently ethical forces, while specialized vocabulary becomes a practical barrier whenever multiple disciplines must solve one problem.
His medical example is “idiopathic thrombocytopenic purpura”: impressive Latin that merely says the patient has low platelets, a rash, and the physician does not know why. Zaks’s career strategy was to learn other functions’ languages; Moderna eventually convinced him that investor expectations were another indispensable language.
6. Great biotech investors study success and preserve optionality
Zaks once favored Mandela’s line, “In life, I’ve either succeeded or I’ve learned.” COVID changed his mind: the quote implies success teaches less than failure, when identifying whether a win came from luck, talent, structure, or judgment may be the harder and more valuable exercise.
His emerging pattern is strong content plus adaptable people. A narrowly staffed, underfunded oncology investment failed without enough scientific, managerial, or financial “degrees of freedom” to change course—a structure poorly suited to a field where progress is almost never linear.
Early Moderna had greater confidence in mRNA’s engineering properties than in its eventual medical market: the molecule could be made repeatedly, reproducibly, and cheaply, but the best use might have been rare disease, oncology, vaccines, or something undiscovered. Parallel exploration preserved that option while capital shifted toward vaccines as the clearest proof of pharmacology.
In December 2019, Stéphane Bancel treated Wuhan as an urgent opportunity while conventional wisdom expected another outbreak to peter out. Zaks’s lesson is not that the outcome was predictable, but that leadership, prior optionality, and control over the marginal dollar made rapid commitment possible.
7. COVID speed was built before anyone heard of COVID
By early 2020, Moderna had used mRNA to generate neutralizing antibodies in humans against eight different viruses, succeeding eight out of eight times. “COVID was to be our ninth”—an unusually strong platform record in a field where individual drugs regularly fail.
The government relationship also predated the crisis. A 2017 Zika collaboration acquainted NIH and BARDA with Moderna’s speed; in September 2019, Zaks told Tony Fauci that mRNA was the best vaccine platform he had seen and explained why. NIH subsequently identified mRNA as a leading pandemic-readiness platform in November 2019.
Moderna and NIH had already planned a timed demonstration: NIH would choose an obscure virus and provide its sequence; Moderna would manufacture a batch; NIH would run phase 1. The real pandemic effectively supplied the test before that exercise began.
Personalized cancer vaccines provided an unexpected manufacturing foundation. Cancer patients required individualized small batches with fast turnaround; that batch size also suited a two-dose phase 1 trial. In Zaks’s telling, COVID benefited from years of cancer-vaccine work, not merely the reverse.
8. Coordination and trial design compressed time without removing proof
Under Operation Warp Speed, NIH, CDC, FDA, and industry settled protocols, endpoints, and trial designs in about a week rather than through the usual weeks or months of exchanges. The pandemic enabled unusually fast decision-making, not a different efficacy standard.
Manufacturing expanded before success was known. Moderna initially raised private capital to fund commercial scale-up; the government later “backstop[ped] the investment” by committing to pay for doses, limiting downside without supplying the original expansion capital.
Phase 3 speed came from event forecasting. Investigators predicted where infections would surge and opened sites in those hotspots, so the event count expected to accumulate over as much as a year arrived within three months.
9. Safety surveillance worked better than public legitimacy
Zaks calls the mRNA vaccines “the most well-studied medical intervention in the history of mankind.” Thousands of people reviewed tens of thousands of reports as exposure expanded from millions to billions; reported event rates ran 10-20 times above prior vaccines because fear, novelty, and new reporting tools drove unusually intensive scrutiny.
The surveillance detected rare events occurring around one in 150,000 cases with adenovector vaccines, prompting rapid labeling and declining use. It also identified rare myocarditis in young adults after mRNA vaccination and added it to the label; Zaks stresses that the virus itself posed a higher myocarditis risk.
His concession is institutional rather than evidentiary. Scientists, “myself included,” assumed science’s benefit justified directing behavior, but science is not an ethical framework. Vaccine mandates balance health against autonomy and freedom; in a democracy, that conflict belongs in public deliberation, not solely to physicians or scientists.
10. Nucleic acids turn medicines into editable information
Zaks broadens the platform thesis from mRNA to nucleic-acid medicines, including siRNA. Once delivery and manufacturing work, the physical construct remains largely constant while encoded information changes the output—“drugs as information,” with each subsequent program costing a fraction of the first.
For rare genetic disease, transient mRNA can supply a missing protein but requires repeated dosing. The company Zaks calls Exilio instead uses mRNA in a lipid nanoparticle to carry information intended to enter the nucleus and make a lasting change.
He draws a sharp distinction: conventional mRNA does not alter DNA, “based on first principles,” but there are ways to encode different information with mRNA that can intentionally make such a lasting change. The latter could replace injections every two weeks with one, two, or three treatments and a lifelong effect.
Permanence creates a new social bargain. Patients may object to intentional DNA alteration, and developers must distinguish changes limited to an individual from intended or unintended effects on offspring. Zaks treats those objections as legitimate ethical questions, not scientific misunderstandings to be dismissed.
11. Healthcare can move from treating illness to improving health
Zaks introduces his forecast with a Talmudic warning that prophecy belongs to fools, then admits he has changed his own mind. He once thought personal monitoring merely repeated advice to eat well and exercise; he now believes population averages “peanut butter” over meaningful individual differences and plans deeper data collection himself.
Routine hospital blood panels remain nearly the same as when he trained 40 years ago, perhaps with two additional analytes. Amos Tanay’s work, using roughly a quarter-million people over a decade as Zaks recalled it, suggested about half the apparent normal variation could be explained by individual factors and corrected—potentially revealing abnormalities earlier.
The unresolved step is actionability: earlier signals matter only if medicine can prove an earlier intervention improves outcomes. Still, vaccines changed Zaks’s definition of medicine—an oncologist tries to restore a patient’s past health, while a vaccine “takes a healthy person and makes them healthier in the future.”
Over 10-30 years, he expects physicians to become translators of machine-generated knowledge rather than repositories of every answer. That could restore a human role eroded as employed doctors lost agency to healthcare-system algorithms, prescribing constraints, shortened visits, and reduced time with patients.
12. Commercial plumbing may constrain medicine more than technology
Technology already exceeds the system’s ability to deploy it. FDA-approved software can read pathology slides more accurately than a pathologist in some applications, Zaks says, yet only a tiny fraction of tissue-microsection slides are read computationally because replacing microscopes, digitizing workflows, and assigning the investment return remain unsolved.
Antibiotics are his warning against assuming useful science naturally attracts capital. The likelihood of dying from significant infections is rising despite available scientific tools; development lacks commercial incentive.
The same danger applies to rare disease and other long-duration programs. Policy determines whether investors can earn enough during the patent period to fund innovations that later become cheap generics for children and grandchildren.
13. Biotech portfolios need conviction without fantasy economics
Software venture can tolerate 99 failures if one company returns 1,000 times capital. Biotech cannot: 3-4x is the target, 5x is excellent, 10x is uncommon, and larger outcomes are rare. Every investment therefore needs someone willing to “pound the table,” but the portfolio cannot rely on belief alone.
The hardest variable is often not whether a drug has some effect, but the magnitude of clinical benefit. In a competitive market, “slightly worse” is dead; the product must differentiate enough to matter.
Zaks invokes the Hoosac Mountain tunnel: experts expected a hard outer shell and soft interior, only to discover solid rock throughout. Once halfway in, they kept digging. The lesson is that projects are usually harder than underwriting suggests, making resilient teams and financing—not optimistic forecasts—the protection.
Executives supply the conviction that portfolio investors cannot. Asked about Moderna’s stock, Zaks replied that he was investing something more valuable than capital: “my time on Earth.” His rule became to place capital behind credible “sweat equity”—people whose talent and lives are committed to making the science work.
14. AI’s bull case is integration; wisdom remains knowing what to ask
Zaks’s AI bear case is slow progress through narrow verticals: one transcription tool for nurses and nursing homes, another system matching patients to trials, each fighting separately for integration and ROI. The precedent is electronic records, whose hospital implementation often reduced productivity despite technology’s supposed purpose.
The bull case realigns incentives around integrated systems that genuinely increase healthcare productivity; adoption could then accelerate quickly. Patrick’s more utopian possibility—digital twins replacing clinical trials and turning biology into engineering—remains for Zaks “a little bit the realm of science fiction,” bounded by patient autonomy and a deliberately messy democracy.
Deep credentials are not mandatory for investors who can identify experts, ask the right questions, and understand why an answer deserves trust. Bancel’s method was the “five whys” followed by plain English; Zaks’s PhD mentor put it more sharply: if you cannot explain the work to a kindergartener, you do not understand it.
That people-first thesis closes with Steve Rosenberg, who supported Zaks long after Zaks felt he had not been one of his best postdocs, then partnered on Moderna’s personalized cancer vaccine roughly 20 years after their 1998 mouse work. Zaks’s response to that quarter-century of belief is to pay it forward when former colleagues call.