AI's Cancer Cure: This Startup Says It Has the Blueprint

Vivodyne, a biotech startup, asserts that the artificial intelligence (AI) drug discovery sector is hampered by a significant data deficiency, a problem they claim to have addressed with a novel machine.

The company's solution, named HIVE, consists of modular robotic laboratories capable of cultivating 20 distinct types of human tissue. These labs autonomously administer dosages and monitor the tissues, thereby producing the crucial causal biological data that current AI models lack. This type of data is presently derived primarily from animal testing or analyses of individual cells and proteins, rather than from living human tissue.

Andrei Georgescu, CEO and co-founder of Vivodyne, critically questions the efficacy of AI models without human testing, stating, "Absent human testing, what are these [AI] models going to do? They’re going to cure cancer in mice."

Echoing this sentiment, Anthropic CEO Dario Amodei recently commented that assertions regarding AI's ability to cure cancer have transitioned from credible prospects to mere clichés, emphasizing that "the thing that will work is actually curing cancer."

It is worth noting, however, that the concept of AI curing cancer has been previously put forth by Amodei himself in past writings. Similarly, Sam Altman has frequently invoked cancer cures to justify OpenAI's pursuit of Artificial General Intelligence (AGI) and expanding computational resources, while Google DeepMind's Demis Hassabis projected last year that AI could eradicate all diseases within a decade.

Despite these ambitious claims, tangible results have remained modest. While a few AI-designed drugs have advanced to human trials, with one reaching Phase III extensive human testing, the fundamental challenges encountered are often not currently surmountable by AI technologies.

AlphaFold, a Nobel Prize-winning innovation, significantly advanced our comprehension of life's fundamental components, yet it has not, to date, yielded a new pharmaceutical drug. Isomorphic Labs, established to leverage AlphaFold's capabilities, anticipates initiating its first trials by the end of this year, ahead of its initial 2025 schedule. In February, the company articulated that genuine drug discovery necessitates "highly accurate predictive models, across an expansive range of biochemical properties and interactions."

Georgescu contends that the field requires "a sanity check," arguing that current models lack the necessary data to accurately represent the intricate complexity of human biology. This predicament mirrors a long-standing challenge within the pharmaceutical industry, where 90% of drugs demonstrating efficacy in animal trials fail to secure regulatory approval for human use.

Vivodyne's approach diverges significantly. Spun out of the University of Pennsylvania in 2021 following Georgescu's completion of a PhD in bioengineering, the company asserts that its cultured tissues faithfully replicate the behavior of genuine human organs. Specifically, its liver cells demonstrate 94% predictive accuracy when compared to human toxicity trials, its airway tissue aligns with real human tissue behavior 96% of the time, and its bone marrow has achieved 100% concordance across tests involving 20 distinct chemotherapy drugs.

Last week, Vivodyne, having secured just under $80 million through two funding rounds spearheaded by Khosla Ventures, inaugurated what it terms the world's most extensive "human data center" near San Francisco. Georgescu highlights that his team is already achieving double the experimental throughput of all animal trials currently conducted in the United States.

The core objective is to expedite the progression of drug candidates by establishing a clearer understanding of their efficacy prior to undertaking costly clinical trials, which often entail expenses in the tens of millions of dollars. While Vivodyne maintains confidentiality regarding its partners, it confirms collaborations with several prominent pharmaceutical companies to address a challenge Georgescu likens to automotive crash testing. He explains that while car manufacturers typically possess high confidence in their vehicles passing NHTSA requirements before physical testing, drug developers seldom share this assurance when commencing clinical trials, where the vast majority of drugs ultimately fail to secure FDA approval.

Beyond immediate applications, a broader vision guides Georgescu: he views his autonomous biology laboratories as pivotal for generating the causal data essential for training advanced AI models on human biology. He references recent research, such as a study published last month in Nature Methods, which indicates an absence of clear data scaling laws when training generative AI models using existing cellular data.

Georgescu elaborated to TechCrunch that "All the training is done on static snapshots of these cells, and the models are not conditioned at all by the how a cell got to that state." He clarified, "In other words, the model learns ‘this is cell state A,’ ‘this is cell state B,’ but never ‘cell state B is the effect of inflaming cell state A.’"

In contrast, Vivodyne's HIVE machines continuously monitor hundreds of thousands of experiments where diseased tissue is subjected to various stimuli. Georgescu anticipates this will furnish the reinforcement learning necessary to develop AI models capable of understanding human biology with sufficient depth to drive substantial advancements in healthcare.

Georgescu posits that this capability will be crucial not only for addressing current medical challenges but also for shaping a future where intricate diseases demand multi-pathway targeted drugs, a significant departure from most contemporary therapeutics.

He further explained to TechCrunch, "If we want combination therapies, the space that has to be searched explodes—it can’t be an experimental approach." He concluded, "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."