Skip to content
a16za16z

Faster Science, Better Drugs

Can we make science as fast as software? In this episode, Erik Torenberg talks with Patrick Hsu (cofounder of Arc Institute) and a16z general partner Jorge Conde about Arc’s “virtual cells” moonshot, which uses foundation models to simulate biology and guide experiments. They discuss why research is slow, what an AlphaFold-style moment for cell biology could look like, and how AI might improve drug discovery. The conversation also covers hype versus substance in AI for biology, clinical bottlenecks, capital intensity, and how breakthroughs like GLP-1s show the path from science to major business and health impact. Timecodes: 00:00 Introduction to Accelerating Science 00:35 Welcome to the Podcast 00:45 The Moonshot: Virtual Cells and Human Biology 01:57 Challenges in Scientific Progress 02:58 Interdisciplinary Collaboration at Arc Institute 05:11 The Role of AI in Biology 10:18 Understanding Virtual Cells 22:13 Biotech and Pharma Industry Insights 27:39 Challenges in Clinical Trials 28:13 Capital Intensity and Technological Advancements 29:02 Improving Drug Development 30:12 Market Impact and Industry Trends 33:19 Future Technological Breakthroughs 38:15 AI in Drug Discovery 45:50 Investment Focus and Future Prospects 54:02 Closing Remarks and Upcoming Initiatives Resources: Find Patrick on X: https://x.com/pdhsu Find Jorge on X: https://x.com/JorgeCondeBio Stay Updated: Find a16z on X: https://x.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Listen to the a16z Podcast on Spotify: https://open.spotify.com/show/5bC65RDvs3oxnLyqqvkUYX?si=3E8B3qT9TyiwAHJ7JnaKbg Listen to the a16z Podcast on Apple Podcasts: https://podcasts.apple.com/us/podcast/a16z-podcast/id842818711 Follow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures.

Patrick HsuguestJorge CondeguestErik Torenberghost
Sep 14, 202555mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Building virtual cells with AI to accelerate drug discovery

  1. Arc Institute’s core bet is that modeling the cell—the fundamental unit of biology—via foundation models can unlock scalable, practical simulation of biology and accelerate experimental cycles.
  2. Scientific progress is slow due to incentives, fragmented multidisciplinary work, and the real-world time required for making and testing biological hypotheses, not just computing predictions.
  3. The “virtual cell” is operationalized as perturbation prediction: given a starting cell state and desired target state, a model proposes interventions (genes, drugs, combinations over time) to move cells across a learned state manifold.
  4. AI’s biggest near-term constraint in biotech is evaluation and ground-truth feedback: biology is harder to “see,” requires lab-in-the-loop validation, and suffers from missing measurements, even if transcriptomics can serve as a scalable mirror for other layers.
  5. Even if AI dramatically improves design and target selection, drug development bottlenecks remain in physical manufacturing, safety/efficacy testing, clinical trials, and regulatory timelines—driving the industry’s capital-intensity challenges.

IDEAS WORTH REMEMBERING

5 ideas

Start with cells before attempting whole-body digital twins.

They argue it’s premature to model entire bodies over time if we can’t reliably predict outcomes in a single cell; accurate cell-level prediction is a more scoped, rigorous stepping-stone toward higher-level biological simulation.

The practical “virtual cell” product is a wet-lab co-pilot, not a benchmark winner.

Arc frames success as generating actionable experimental plans—e.g., which 12 perturbations to run next—rather than marginal improvements on abstract ML metrics like gene-expression error.

Biology AI is slowed by evaluation, not just modeling.

Unlike text or images, we don’t natively interpret DNA/cell outputs; progress requires lab-in-the-loop ground truth and better ways to interpret “fuzzy” model outputs against experimental reality.

Transcriptomics can be a scalable proxy even if it’s incomplete.

They acknowledge missing modalities (metabolites, spatial and temporal dynamics) but claim RNA at scale can reflect protein/metabolic signaling indirectly, enabling early capability while richer measurements mature.

Drug failures cluster into two buckets: wrong target or wrong molecule.

With ~90% clinical trial failure rates, they emphasize virtual cells could improve target ID and perturbation planning, but new “drug matter” is still needed for tissue-specific or pleiotropic targets.

WORDS WORTH SAVING

5 quotes

I want to make science faster. Our moonshot is really to make virtual cells at Arc and simulate human biology with foundation models. Why are we so worried about modeling entire bodies over time when we can't do it for an individual cell?

Patrick Hsu

It's this weird Gordian knot that ultimately comes down to incentives, right?

Patrick Hsu

We don't speak the language of biology, right? You know, you know, at, at very best with an incredibly thick accent, right?

Patrick Hsu

If we have ninety percent of drugs failing clinical trials, right, that kind of means two things, and you're not sure what percent of which, right? One is we're targeting the wrong target in the first place. The second is the composition, the drug matter that we're using doesn't do the job, right?

Patrick Hsu

The, the, the crazy thing is the progress in just the short time that I've been doing this is insane... at Arc, in the next, you know, kind of n, like, I don't know, relatively short amount of time, we're gonna generate a billion perturbed single cells, right? That's like, I mean, how's, how's that for a Moore's Law?

Patrick Hsu

Why science is slow: incentives, training, fragmentationArc Institute’s “collision frequency” organizational modelVirtual cells and perturbation prediction frameworkScaling laws and “RNA as a mirror” for hidden biologyAlphaFold’s impact vs limits for drug discoveryClinical trial bottlenecks and 90% failure rateBiotech business models, capital intensity, and GLP-1 market signal

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome