Scuttlebutt

Scuttlebutt

Virtual cell models stop improving after a couple percent of their data — Vivodyne CEO

6 insights: why cancer cell therapies that kill tumors in a dish miss them in patients, how a robot compiles an experiment, and why common drug toxicity matters more than rare cases

Scuttlebutt's avatar
Scuttlebutt
Sep 28, 2026
∙ Paid

The Cognitive Revolution · Sep 27, 2026 · 20m · 10m saved

With Andrei Georgescu, Co-Founder and CEO of Vivodyne

Virtual cell models stop improving after a couple percent of their data — Vivodyne CEO

The most advanced virtual cell models stop improving after training on a couple percent of the data they are given, Andrei Georgescu said.

The usual objection to AI in biology is that nobody can ever collect enough data. Georgescu's objection is to the kind of data: cells growing in a dish behave nothing like cells in a body, so more of it teaches a model very little.

"So you can have all the data you want, but their performance saturates after a couple percent."

Andrei Georgescu, co-founder and CEO of Vivodyne, on The Cognitive Revolution, runs robotic labs that grow lung, liver, tumor and other human tissue from primary human cells and test drugs on it. In August the company opened 12 of those labs, which by its count can grow more than 3 million tissues a year.

The full segment is covered here so you can skip it. 20 minutes of audio, 10 minutes of reading.

Here are the 6 insights that matter.


This Scuttlebutt Summary is available in full to premium subscribers. If you're not one already, please consider subscribing to keep reading.


Key Takeaways

  • Cell therapies that kill tumors in a dish can flow right past them in patients

    • Vivodyne doses drugs through blood vessels the tissue grows itself

  • Virtual cell models saturate after a couple percent of their data, because cells in a dish only proliferate

  • Vivodyne starts from mature human cells, not stem cells, and lets them rebuild the tissue's structure

  • The model decides which experiments to run; the lab checks whether it was right

  • Its second-generation tissue disk holds two to four times as many tissues, thanks mostly to software

  • Common drug toxicity is a bigger problem than rare adverse events, Georgescu said

1. Dosing Through Blood Vessels

The host said Vivodyne builds robotic labs that grow pieces of human tissue from primary human cells and tests drugs on them. The co-host asked how Vivodyne works out a dose, and whether it matches what a cell would meet in the body.

Georgescu said nearly all preclinical testing outside animals soaks an organoid, or cells in a petri dish, in the test compound. That is not how drugs move through the human body.

The CAR-T example

"A lot of cell therapies that are designed for solid tumors, in a dish will completely kill that tumor. And then those same CAR T cells, when they go into a patient, just flow right by the tumor through the blood vessels. They don't even know it's there." — Andrei Georgescu

The challenge, he said, is modeling how a therapy travels into the tissue: whether the cells recognize the tumor and whether they concentrate there.

The tissue grows its own capillaries

In Vivodyne's tissues, the small blood vessels assemble themselves, and fluid can be pumped through them.

"And whenever we dose a drug, it is always through the blood vessels that are native to that tissue." — Andrei Georgescu

The tissue beyond the vessel walls has the same density as normal tissue, he said, so whether a drug reaches its target counts as much as whether it binds to it.

2. Virtual Cells Hit a Ceiling

User's avatar

Continue reading this post for free, courtesy of Scuttlebutt.

Or purchase a paid subscription.
© 2026 The Golden Age LLC · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture