a16zTaking Bold Bets: NIH and the Future of Biomedical Science
CHAPTERS
Why public trust depends on evidence, transparency, and productive failure
Bhattacharya opens with a thesis about rebuilding trust: treat the public as intelligent partners, show the data, and allow disagreement. He argues science should adopt a “Silicon Valley spirit” that rewards learning—including learning from failure—rather than punishing it.
- •Trust grows when the public is shown evidence directly, not asked to defer to authority
- •Productive failure should be publishable and valued as learning
- •Scientific culture should encourage bold bets rather than penalize risk
- •Respectful debate is framed as a prerequisite for legitimacy
Autism Data Science Initiative: $50M, 13 teams, and a push for causes & prevention
He announces a new NIH autism effort prompted by rising prevalence and lack of clear answers for families. The initiative funds large, data-science-oriented projects intended to accelerate understanding of causes and potential prevention strategies.
- •Autism prevalence has risen for decades; families lack clear causal explanations
- •$50M new funding; ~250 teams applied; 13 teams selected
- •Goal: accelerate rigorous, data-driven research that can yield actionable answers
- •Motivation includes improving therapies and informing prevention, not just description
Clinical signals in focus: leucovorin for subsets of autism and caution on Tylenol in pregnancy
Two additional announcements highlight translation to care: expanding access to leucovorin (folinic acid) for specific folate-processing issues, and updated caution/guidance around acetaminophen use during pregnancy. He emphasizes nuance—potential benefits for some, uncertainty for others—and avoiding panic while acting prudently.
- •Leucovorin can help a subset of autistic children with folate processing problems; some see speech and symptom improvements
- •Not universal—requires the right biological context; stresses targeted use
- •Emerging evidence suggests correlation between acetaminophen in pregnancy and later autism diagnoses; controversy remains
- •FDA working on revised guidance; CMS working on payment/coverage changes for leucovorin
Preterm birth as a national outcomes gap and a family-centered research mandate
The conversation expands to preterm birth, noting the U.S. has worse outcomes than parts of Europe and lacks sufficient explanations. Bhattacharya frames NIH’s role as delivering rigorous science that responds to real family concerns, while acknowledging complexity and multiple contributing factors.
- •U.S. preterm birth outcomes lag Europe; causes are multifactorial
- •Prenatal care access matters but is not the whole story
- •NIH’s job is to produce rigorous science that can translate to clinical insight
- •Emphasizes listening to public concerns as a driver of research priorities
The replication crisis: why “published” doesn’t mean “true”
Bhattacharya diagnoses replication as the core standard for scientific truth and argues current incentives undervalue verification. He attributes the crisis to the difficulty and scale of modern science, specialization, and weak incentives to check others’ work or publish negative/failed replications.
- •Replication by independent teams should be the standard for truth
- •Volume and specialization make cross-checking harder than in earlier eras
- •Career incentives reward novelty and publication, not verification or replication
- •Peer review and journal publication are treated as overly strong signals of correctness
NIH reform agenda: auditability, centralized review, and modernizing collaboration oversight
Reflecting on his first months, he highlights operational reforms: strengthening accountability for foreign collaborations and centralizing grant review processes. He argues these changes preserve global collaboration while ensuring the U.S. can track funds and verify compliance.
- •Foreign collaborations should continue, but must be auditable (e.g., lab notebooks, financial tracking)
- •Responds to criticisms/misreadings that oversight equals ending collaborations
- •Centralized review via the Center for Scientific Review to reduce parallel, inconsistent systems
- •Goal: improve integrity, comparability, and confidence in NIH funding decisions
Bringing venture-style portfolio thinking to NIH: bold bets, tolerance for failure, and newer ideas
He draws a direct analogy to venture capital portfolios: many failures can be acceptable if the portfolio produces transformative wins. He argues NIH became more conservative over time—funding older, safer ideas—and needs to renew incentives for experimentation and learning.
- •Portfolio logic: success is measured across the whole set, not by every individual project
- •NIH shifted from funding very new ideas (’80s/’90s) to older ideas (2000s/2010s)
- •Peer review can suppress disruptive ideas that challenge incumbents
- •Calls for cultural change: tolerate and learn from productive failure
Allocation vs execution: who decides priorities, and why politics can’t be removed
Agarwala frames NIH’s challenge as both allocating dollars across areas and executing well within each area; Bhattacharya agrees and explains why public (political) input is appropriate for macro allocation. He argues scientists are essential within-area decision makers, but not legitimate “philosopher kings” for societal tradeoffs.
- •Macro allocation reflects both public will and scientific opportunity; Congress and the President set budgets
- •Scientists guide within-area strategy and opportunity evaluation
- •Public/patient activism (e.g., early HIV) can correct scientific under-response
- •Tradeoffs across diseases are value-laden and require democratic legitimacy, not only expertise
Early-career investigator bottleneck: aging of first grants, postdoc inflation, and new incentives
Bhattacharya describes how the system increasingly delays independence for young scientists, shifting first major grants from mid-30s to mid-40s and encouraging multiple postdocs. He proposes changes that judge institutes on portfolio outcomes, strengthen alignment to strategic plans, and reward mentorship and early-career support.
- •Median age for first major NIH grant rose from ~35 to mid-40s; long postdoc chains are common
- •New ideas skew younger; idea “novelty” tends to age with the scientist unless actively resisted
- •Plan: empower institute directors to manage portfolios (not punish individual failures)
- •Incentivize funding aligned to strategic plans and improve support/mentorship for early-career investigators
Training and the “missing link”: bridging from training grants to faculty independence
The discussion turns to NIH training mechanisms (pre-doc, postdoc, MSTP) and where attrition occurs. Bhattacharya argues the biggest failure point is the transition from training to independent positions and that NIH should reward institutions and structures that make that leap feasible.
- •NIH supports talent early (undergrad to postdoc), but many still drop out later
- •Key gap: post-training transition to independent assistant professor roles (e.g., difficult K awards)
- •Universities and system incentives can prolong dependency and delay independence
- •Policy emphasis: improve the pathway to independence rather than only earlier-stage support
Rebuilding public health trust after the pandemic: gold-standard science and humility
Bhattacharya links today’s trust deficit directly to pandemic-era policies he views as weakly evidenced and harmful. He proposes two pillars for repair: enforce “gold-standard science” norms (replication, unbiased review, humility) and reframe public health as servant-partner to the public rather than a commanding authority.
- •Pandemic examples cited: plexiglass, inconsistent masking rules, school closures and downstream harms
- •Trust repair requires rigor (replication, unbiased processes) and transparent limits of evidence
- •Public health should act as partner/servant, not top-down enforcer
- •Acknowledges rebuilding trust will take a long time
Communicating uncertainty: say ‘I don’t know,’ avoid false certainty, and invite open debate
He argues that in uncertain situations, honesty about unknowns is essential and that overconfident guidance erodes credibility. At the same time, he distinguishes between areas of genuine uncertainty and areas with strong evidence (e.g., MMR), advocating open scientific discourse rather than censorship or “canceling.”
- •Best practice in uncertainty: explicitly state unknowns and describe how answers will be sought
- •Pandemic-era error: projecting certainty without adequate evidence
- •For well-established interventions (e.g., MMR), communicate strength of evidence while staying open to challenge
- •Academic freedom and public reasoning are positioned as the mechanism for resolving disputes
NIH priorities: chronic disease, nutrition, and AI—plus practical examples like Alzheimer’s prevention signals
Bhattacharya outlines priorities including chronic disease burden and integrating AI, emphasizing translational opportunities. He highlights observational findings suggesting an old shingles vaccine (Zostavax) may reduce Alzheimer’s cognitive decline risk, illustrating the kind of pragmatic, high-upside hypothesis he wants to pursue.
- •Chronic disease focus tied to stagnant U.S. life expectancy and growing incidence burdens
- •Example: observational work suggesting Zostavax associated with reduced Alzheimer’s cognitive decline risk
- •Emphasis on finding simple, scalable interventions when evidence supports testing them
- •AI is framed as an accelerator for discovery and care delivery, contingent on careful validation
The role and limits of AI: augmentation over substitution, plus guarding against system noise
AI’s promise includes protein-structure tools (e.g., AlphaFold), radiology assistance, and reducing clinician documentation burden. But Bhattacharya warns of hallucinations, low-quality AI-generated grant spam, and the need for policies that keep AI from overwhelming review systems while enabling secure internal use.
- •AI can accelerate drug discovery and focus wet-lab work; improves clinical workflows (e.g., documentation)
- •Need rigorous research/validation to prevent hallucination-driven harms
- •NIH/HHS deploying secure AI tools; privacy and security are emphasized
- •Policy response to AI-generated “noise”: limit number of applications per cycle and discourage mass low-quality submissions
Advice to scientists: persist like Max Perutz—and why portfolios must allow long, uncertain bets
Closing advice centers on resilience and conviction: transformative discoveries often require years of persistence despite skepticism. He uses Max Perutz’s decade-long pursuit of protein structure to argue NIH must create conditions where such long-horizon, high-risk work can survive and succeed.
- •Science has outsized potential to improve human well-being; individuals drive breakthroughs
- •Persistence through repeated rejection is portrayed as a key trait of impactful scientists
- •Max Perutz story illustrates long gestation periods for paradigm-shifting work
- •Reinforces portfolio philosophy: invest across approaches (molecules, care management, lifestyle) because outcomes are hard to predict
