Vivodyne, a biotech startup spun out of the University of Pennsylvania, argues that AI-driven drug discovery is held back by poor or missing human biological data. The company has developed modular robotic laboratories called HIVE that can grow 20 types of human tissue, autonomously dose and monitor them, and produce longitudinal, causal biological datasets that Vivodyne says are largely absent from current model training regimes. Today’s datasets, the company notes, mainly come from animal experiments or from single-cell and protein studies rather than from living human tissue.
The problem with current data
Andrei Georgescu, Vivodyne’s CEO and co-founder, warns that without human testing AI models risk learning misleading patterns: “Absent human testing, what are these [AI] models going to do? They’re going to cure cancer in mice,” he told TechCrunch. Similar skepticism has been voiced by others in the field; Anthropic CEO Dario Amodei wrote that claims AI will cure cancer are increasingly clichéd rather than credible.
The track record so far is mixed. Some AI-designed drug candidates have entered human trials — one reaching Phase III — but many obstacles remain beyond what current AI can resolve. Breakthroughs like AlphaFold advanced understanding of protein structure, yet have not directly yielded new drugs. Companies building on such advances, such as Isomorphic Labs, have delayed expectations: Isomorphic originally targeted 2025 for first trials and said in February that genuine drug discovery will require “highly accurate predictive models, across an expansive range of biochemical properties and interactions.”
Vivodyne’s data and claims
Vivodyne was formed in 2021 after Georgescu completed a PhD in bioengineering at the University of Pennsylvania. The company reports high concordance between its tissues and human outcomes in certain tests:
- Liver cells demonstrated 94% predictive accuracy compared with human toxicity trials.
- Airway tissue models matched real human tissue behavior 96% of the time.
- Bone marrow models showed 100% concordance across tests of 20 different chemotherapy drugs.
Financially, Vivodyne has raised just under $80 million across two funding rounds led by Khosla Ventures. Last week the company opened what it describes as the world’s largest “human data center” just outside San Francisco. Georgescu says the facility is already achieving twice the throughput of all animal trials currently run in the U.S.
How this could change preclinical screening
The stated goal is to speed selection of drug candidates by giving developers a better sense of what is likely to work before entering costly clinical trials, which typically run into tens of millions of dollars. Vivodyne will not name its pharma partners publicly, but says it is collaborating with multiple major drugmakers to address a problem Georgescu compares to automotive crash testing: automakers typically have high confidence a vehicle will meet NHTSA requirements before testing it, while drugmakers rarely have comparable confidence before starting clinical trials, where most drugs fail to gain regulatory approval.
Aiming to provide causal datasets for AI
Beyond preclinical screening, Georgescu sees HIVE’s output as critical training data for next-generation AI models that must understand causal relationships in human biology. He cites recent research published in Nature Methods that found no clear data-scaling laws when training generative AI on existing cellular datasets. According to Georgescu, most training is done on static snapshots of cell states, so models learn labels like “this is cell state A” and “this is cell state B,” but not the causal processes by which A becomes B.
Vivodyne’s HIVE systems track hundreds of thousands of ongoing experiments where diseased tissue is exposed to stimuli and the subsequent dynamics are recorded. Georgescu expects those longitudinal, intervention-conditioned datasets to enable reinforcement learning approaches that could make AI models better at predicting human biological responses.
He also argues such causal understanding will be essential for future combination therapies, where multiple pathways must be targeted simultaneously. “If we want combination therapies, the space that has to be searched explodes—it can’t be an experimental approach,” he told TechCrunch. “You have to say, ‘I want this effect to happen, so what cause should I invoke?’ Establishing causality in human biology is the basis of all of this.”
Limits and open questions
While the accuracy figures and the new facility are notable, Vivodyne has not disclosed partner names and independent validation of whether these models and datasets will meaningfully reduce clinical failure rates remains to be seen. Improving preclinical models is important but not necessarily sufficient to guarantee human success; the central question is how well datasets generated from lab-grown human tissues and automated experiments will generalize into reliable, clinically predictive AI models.
Vivodyne’s proposition is that by delivering higher-fidelity, causal human data, AI can be moved closer to producing medically meaningful advances. The next several years will show whether that data is enough to substantially increase the speed and success rate of drug discovery.
