Key facts
- Vivodyne, a biotech startup, states that current AI drug discovery models lack sufficient causal biological data.
- The company has developed HIVE, robotic labs designed to grow and monitor 20 types of human tissue.
- Vivodyne claims its tissue models achieve high predictive accuracy for toxicity and chemotherapy drug efficacy.
- The startup opened a new "human data center" and aims to accelerate drug development by improving pre-clinical predictions.
- Vivodyne is collaborating with major pharmaceutical companies to address data limitations in AI drug discovery.
Vivodyne, a biotech startup, is challenging the current approach to AI-driven drug discovery, arguing that the industry's primary issue is a lack of high-quality, causal biological data rather than the size of AI models.
The company has developed HIVE, a system of modular robotic labs capable of growing 20 different types of human tissue. These labs autonomously dose and monitor the tissues, generating data that Vivodyne claims is crucial for training effective AI models, unlike current methods that often rely on animal testing or studies of single cells.
Vivodyne CEO Andrei Georgescu stated that existing AI models are limited because they lack the complexity of human biology, leading to high failure rates in clinical trials, where 90% of drugs effective in animal tests do not receive human approval. He likens the current situation to automotive safety testing before modern crash simulations.
Vivodyne, spun out of the University of Pennsylvania in 2021, reports that its tissue models demonstrate significant predictive accuracy. Its liver cells show 94% accuracy for human toxicity tests, airway tissue matches real human tissue behavior 96% of the time, and bone marrow achieved 100% concordance in tests for 20 chemotherapy drugs.
Last week, Vivodyne opened what it calls the world's largest "human data center" near San Francisco, having raised nearly $80 million from investors like Khosla Ventures. The company asserts its throughput is double that of all animal trials conducted in the U.S. and is working with multiple major pharmaceutical companies.
Georgescu envisions these autonomous biology labs as essential for generating causal data that can train new AI models to better understand human biology. He points to research indicating that current cellular data lacks scaling laws for generative AI, and that models learn static states rather than cause-and-effect relationships. Vivodyne's approach aims to provide this missing causal link, which Georgescu believes is vital for developing future combination therapies for complex diseases.