Sep 15, 2025 · 55m · a16z
Faster Science, Better Drugs
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Patrick Hsu, Co-Founder of the Arc Institute, joins the a16z Podcast to discuss how artificial intelligence and virtual cells can accelerate scientific discovery and drug development. He examines structural bottlenecks in academia and biotech, separates real AI tools from hype, and outlines future frontiers in synthetic biology and robotics.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The host holds 5.7% of the talking time here. How this is scored →
The host as informed peer 4.8 Guest teaching 4.1 Guest disagreement 1.7 The host pushing back 3.0
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 5:21 Language vs biology modeling difficulty hot take
Patrick directly offers a hot take rejecting the assumption that biology ML is on par with text/vision, claiming natural language modeling is vastly easier because humans already natively speak it.
Hardest push from the host ▶ 43:51 Epistemological critique of biological data completenessJorge directly challenges the fundamental premise of virtual cell models by asking if scientists are feeding models fundamentally incomplete data without knowing what core variables are missing.
Biggest teaching moment ▶ 7:13 RNA transcriptomics as a mirror statePatrick reframes the host's doubt about unobservable cellular components by explaining how massive transcriptomic data acts as a lower-resolution mirror state for underlying protein state changes.
The host holds their own ▶ 25:49 Detailed breakdown of biotech venture economicsCo-host Jorge Conde steps in to articulate the structural realities of biotech investing, outlining clinical trial fail modes, capital intensity, and why regulatory timelines resist AI compression.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| a16z Podcast Title Sequence | 3 | 3 | 1 | 2 | The hosts introduce Patrick Hsu and inquire about the Arc Institute model. Jorge Conde gently probes why traditional universities fail to bring disciplines together under one roof, leading Patrick to explain physical distance and academic incentive structures. | |
| Why AI in Biology Lags Behind Text and Vision | 3 | 5 | 2 | 3 | Patrick points out that AI in biology is harder because humans do not natively speak DNA or cellular language. Jorge presses on how virtual cell models can succeed when unknown biological variables exist, and Patrick explains scaling laws and RNA transcriptomics as a mirror state. | |
| Fleshing Out the Virtual Cell and AlphaFold Moments | 4 | 5 | 2 | 3 | Patrick outlines perturbation prediction and the path toward a virtual cell AlphaFold moment. When Jorge asks if AI discovery will prove biology textbooks wrong, Patrick reframes textbooks as compressed representations rather than incorrect facts. | |
| Virtual Cells vs. Digital Twins: The Right Level of Abstraction | 2 | 4 | 2 | 1 | Patrick dismisses hype terms like digital twins or avatars as media-friendly fluff, arguing that virtual cells represent a more mathematically rigorous and appropriately scoped level of abstraction. | |
| Biotech Business Models and Clinical Trial Bottlenecks | 8 | 3 | 1 | 4 | Co-host Jorge Conde demonstrates deep domain expertise as a biotech investor, detailing clinical trial bottlenecks, capital intensity, and valuation step-up dynamics that AI cannot easily compress. | |
| The GLP-1 Revolution and Expanding Industry Ambition | 7 | 3 | 1 | 2 | Jorge articulates the multi-modal trajectory of drug development across small molecules, biologics, and gene editing. Patrick brings in geopolitical framing around US regulatory law vs Chinese engineering focus. | |
| Hype, Hope, and Heft in AI Drug Discovery | 6 | 4 | 2 | 5 | Jorge reframes Erik's general question into a clear Hype, Hope, and Heft taxonomy. Patrick categorizes toxicity prediction as hype, protein engineering as heft, and virtual cell integration as hope. | |
| Simulated Discovery Agents and Incomplete Biological Data | 7 | 5 | 3 | 6 | Jorge challenges the core premise of virtual cell models by asking what happens if input datasets are fundamentally incomplete. Patrick concedes the point and uses weather forecasting to illustrate mechanistic vs predictive simulation. | |
| Frontiers of AI Investment: Synbio, BCIs, Robotics, and Architectures | 3 | 5 | 1 | 1 | Patrick presents his broader AI investment thesis spanning synthetic biology, brain-computer interfaces, robotics, and post-transformer model architectures like Sakana AI. |
Statements from this episode (22)
Hsu: Arc Institute's moonshot is building virtual cells using foundation models
“Our moonshot is really to make virtual cells at Arc and simulate human biology with foundation models, and, you know, we'd like to figure out something that feels useful for experimentalists, people who are skeptical about technology, you know, they just want …”
Hsu: Research groups struggle to excel at more than two disciplines
“It's very hard for individual research groups or individual companies to be good at more than two things, right?”
Hsu: Academic incentives actively discourage cross-disciplinary research collaboration
“Folks have their own incentive structures, right? They need to publish their own papers. They need to do their own thing and, you know, make their own discovery and, You're not really incentivized to work together.”
Hsu: Arc Institute focuses on Alzheimer's targets and virtual cells
“We have two flagship projects, one trying to find Alzheimer's disease drug targets, the other two make these virtual cells”
Hsu: Modeling natural language and video is easier than modeling biology
“Natural language and video modeling is easier than modeling biology.”
Hsu: Virtual cell models will advance from single cells to whole animals
“There are going to be different phases of capability where initially they model individual cells, then they model pairs of cells, then they model cells in a tissue, and then in a broader physiologically intact animal environment, and those are length scales an…”
Hsu: Scaling single-cell screens alone will feel dated in three years
“We weren't just going to be an engineering shop that's just trying to scale single cell perturbation screens, right? That You know, would be interesting, but in three years would feel very dated, I think, right?”
Hsu: AI pharma commercial pitches outpace fundamental research breakthroughs
“I think in many ways the kind of pitch and the framing of these companies precedes The fundamental research capability breakthroughs.”
Hsu: Current AI biological models are at GPT-1 or GPT-2 capability levels
“I find it helpful to frame these in terms of like GPT one, two, three, four, five capabilities, right? And I think most people would agree we're somewhere between GPT one and two, right?”
Hsu: Virtual cell modeling is more rigorous than digital twin concepts
“I think virtual cells, if anything, is actually way more scoped and rigorous than modeling a digital twin or avatar.”
Hsu: Cell responses to chemical and environmental perturbations will be predictable
“Obviously, we want to be able to predict drug toxicity. We want to be able to predict aging. We want to be able to predict why a liver cell becomes cirrhotic when you repeatedly challenge it with ethanol molecules or whatever, right? And, you know, the, these …”
Hsu: Arc Institute will soon generate one billion perturbed single cells
“And at ARC in the next, you know, kind of end, like, I don't know, relatively short amount of time, we're going to generate a billion perturbed single cells, right?”
Conde: AI has not yet compressed biotech clinical trial timelines
“We still haven't seen, and I think it's coming, but it hasn't happened yet. We haven't seen artificial intelligence or other technologies massively compress the amount of time it takes us to do the clinical development, the clinical trials, the enrollment of p…”
Hsu: GLP-1 market value exceeds 40 years of biotech startups combined
“The first is the amount of market cap added to Lilly and Novo based on the, you know, development of GLP ones. It's like over a trillion dollars is, is more, you know, I mean, NOAA stock has decreased a lot. So you know, trillion dollars, let's say, is more th…”
Hsu: Physical synthesis and testing will remain AI drug discovery's bottleneck
“If we play this out, right? And let's say these AI models work, right? And you can make a trillion binders in silico that will, you know, be exquisite drug matter, right? We still need to make these things physically and test them in animals and hopefully pred…”
Hsu: AI for protein design and binding has real-world heft
“There's heft in anything to do with proteins, right? Obviously protein binding, but increasingly in protein design, right? And I think there's real heft there.”
Hsu: Multimodal biological models in AI biotech are currently mostly hype
“And then, you know, where there's hype is in Multimodal biological models, whatever that means, right?”
Hsu: AI will become a native part of drug discovery in years
“I feel like in, you know, I mean, increasingly in just a few years, this will just be a native part of the stack, right? Just like we use You know, the internet and we use phones. We're going to have AI in all parts of the stack, right? And so it's just going …”
Hsu: High-throughput biology is currently limited to imaging and sequencing
“Like, in biology, we ultimately have two ways to study it in high throughput. It's imaging and sequencing, right?”
Hsu: Virtual cell models do not need mechanistic explanations to be useful
“And I would say similarly with a virtual cell model, it may not tell me literally why, just like a alpha fold doesn't tell me literally why did the protein fold this way and how, but it just told me the end state and it was reasonably accurate. I think that wo…”