• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    ‘Virtual Biotech’ reportedly uses more than 37,000 AI agents to find targets and forecast trial success

    A post sharing a Science release says the clinical trial forecasts addressed both efficacy and avoiding side effects.

    Furong HuangFH
    James ZouJZ
    Eric TopolET
    11 Sources, ,

    TLDR

    A post sharing a Science release describes “Virtual Biotech” as putting more than 37,000 AI agents to work finding targets and forecasting success in clinical trials. The post says that forecasting covered both efficacy and avoiding side effects.

    Combined views

    100.3K

    11 Sources, first seen 20d ago

    Combined views

    100.3K

    11 Sources, first seen 20d ago

    985 likes
    20d ago
    first seen 20d ago
    985 likes
    35 comments
    599 saves
    195 reposts
    35 comments
    599 saves
    195 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    11 Sources

    Eric Topol@EricTopolFirst release @ScienceMagazine "Virtual Biotech," whereby >37,000 AI agents were put to work to find targets and forecast success in clinical trials, both for efficacy and avoidance side-effects. And that was one the first of a series of accomplishments @james_y_zou @harrison_zhang https://www.science.org/doi/10.1126/science.aeg677920d
    James Zou@james_y_zouSuper excited that #VirtualBiotech is published in @ScienceMagazine today! Tens of thousands of AI scientist agents work together across drug discovery + development, from target discovery to clinical trial design. Our blueprint for agentic R&D 🧵20d
    Furong Huang@furonghThis is the power of ultra-scale agent fleets—carefully orchestrated and coordinated! An exciting early glimpse of what multi-agent systems can achieve when done right.20d
    Harrison G. Zhang@harrison_zhang🔥 Beyond thrilled that the Virtual Biotech, our multi-agent AI system for drug discovery, is published in @ScienceMagazine! Discovering a new medicine is one of the most consequential endeavors in biomedicine: a breakthrough can mean years of life, independence, and hope for patients and their families. Yet 90%+ of drugs that enter clinical trials still fail. Drug discovery is intrinsically hard bc it spans an extraordinary range of disciplines, and a successful medicine must check the right boxes. No single person can be an expert in all of these domains, yet decisions in one directly determine success or failure in another. Making better decisions across this immense complexity is how we think AI agents will transform this field. We introduce the Virtual Biotech, an organization of AI scientists modeled on a biotech company. A virtual Chief Scientific Officer coordinates their work, while a scientific reviewer challenges conclusions and identifies missing evidence. Each AI scientist is domain-specialized and equipped with custom tools and knowledge to conduct research in its field. First, more than 37,000 agents assembled and analyzed a dataset of nearly 56,000 clinical trials, finding that drugs targeting cell-type-specific genes were associated with a 48% greater likelihood of successfully reaching market and 32% lower adverse event rates. The system then performed target validation to autonomously build a preclinical rationale for B7-H3 target in lung cancer after analyzing genetics, single cell, spatial transcriptomic, and clinical data. Finally, it conducted a translation failure analysis to flag possible failure mechanisms for a terminated trial testing first-in-class mAb against OSMR in ulcerative colitis. This work points toward a future where every scientist has the breadth of a biotech research organization at their fingertips—significantly expanding what they can imagine, investigate, and ultimately translate into new medicines. We owe it to our patients to make that future a reality. 🤖 Talk to your Virtual Biotech research team: https://virtualbiotech.ai/ 🖇️ Paper: https://www.science.org/doi/10.1126/science.aeg6779 💻 Code: https://github.com/harrisongzhang/TheVirtualBiotech @james_y_zou Peter Eckmann @Jiacheng_Miao Andrew Mahon @StanfordMed @StanfordDBDS @StanfordAILab @KnightHennessy @StanfordHAI @StanfordMSTP19d

    11 Sources

    Eric Topol@EricTopolFirst release @ScienceMagazine "Virtual Biotech," whereby >37,000 AI agents were put to work to find targets and forecast success in clinical trials, both for efficacy and avoidance side-effects. And that was one the first of a series of accomplishments @james_y_zou @harrison_zhang https://www.science.org/doi/10.1126/science.aeg677920d
    James Zou@james_y_zouSuper excited that #VirtualBiotech is published in @ScienceMagazine today! Tens of thousands of AI scientist agents work together across drug discovery + development, from target discovery to clinical trial design. Our blueprint for agentic R&D 🧵20d
    Furong Huang@furonghThis is the power of ultra-scale agent fleets—carefully orchestrated and coordinated! An exciting early glimpse of what multi-agent systems can achieve when done right.20d
    Harrison G. Zhang@harrison_zhang🔥 Beyond thrilled that the Virtual Biotech, our multi-agent AI system for drug discovery, is published in @ScienceMagazine! Discovering a new medicine is one of the most consequential endeavors in biomedicine: a breakthrough can mean years of life, independence, and hope for patients and their families. Yet 90%+ of drugs that enter clinical trials still fail. Drug discovery is intrinsically hard bc it spans an extraordinary range of disciplines, and a successful medicine must check the right boxes. No single person can be an expert in all of these domains, yet decisions in one directly determine success or failure in another. Making better decisions across this immense complexity is how we think AI agents will transform this field. We introduce the Virtual Biotech, an organization of AI scientists modeled on a biotech company. A virtual Chief Scientific Officer coordinates their work, while a scientific reviewer challenges conclusions and identifies missing evidence. Each AI scientist is domain-specialized and equipped with custom tools and knowledge to conduct research in its field. First, more than 37,000 agents assembled and analyzed a dataset of nearly 56,000 clinical trials, finding that drugs targeting cell-type-specific genes were associated with a 48% greater likelihood of successfully reaching market and 32% lower adverse event rates. The system then performed target validation to autonomously build a preclinical rationale for B7-H3 target in lung cancer after analyzing genetics, single cell, spatial transcriptomic, and clinical data. Finally, it conducted a translation failure analysis to flag possible failure mechanisms for a terminated trial testing first-in-class mAb against OSMR in ulcerative colitis. This work points toward a future where every scientist has the breadth of a biotech research organization at their fingertips—significantly expanding what they can imagine, investigate, and ultimately translate into new medicines. We owe it to our patients to make that future a reality. 🤖 Talk to your Virtual Biotech research team: https://virtualbiotech.ai/ 🖇️ Paper: https://www.science.org/doi/10.1126/science.aeg6779 💻 Code: https://github.com/harrisongzhang/TheVirtualBiotech @james_y_zou Peter Eckmann @Jiacheng_Miao Andrew Mahon @StanfordMed @StanfordDBDS @StanfordAILab @KnightHennessy @StanfordHAI @StanfordMSTP19d