Alex Zhavoronkov

As the discourse shifts from models to compound AI systems / agents, we need better AI benchmarks to evaluate multi-modal and multi-step task performance, especially in healthcare and life sciences. When we wrote the first paper demonstrating ChatGPT's performance on the USMLE, we chose the US Medical Licensing Exam as a benchmark for accessibility, speed, and ease. This benchmark was never intended to represent AI model performance on real-world clinical tasks. Today, I still see so many research teams and startups using benchmarks (like the USMLE) that are ill-fitted for assessing the true clinical or scientific performance and utility of the models they are developing for real-world contexts. Benchmark development may be seen as a "less sexy" area of research, but it is of paramount importance. Years after the rise of the transformer, we still lack adequate benchmarks for so many single-step tasks in biomedicine. With compound AI systems (i.e., architectures that integrate multiple AI models to perform complex tasks) emerging, we need new benchmarks for agentic behaviors. I'd argue that developing an agent with novel capabilities without at least proposing a companion benchmark (if an industry standard does not yet exist) may hinder the adoption of said agent, especially for high-stakes workflows. Designing more benchmarks that capture/simulate real-world clinical and scientific workflows will help us mitigate the major discrepancies observed between in silico and in vivo performance and better support safe + effective deployment of AI in biomedicine. There are already brilliant people focused here, and we need more. DMs are open if you're researching or working in this area of multi-step/multi-modal benchmarking in healthcare and life sciences! #healthcare #ai #artificialintelligence #generativeai

100

18 Comments

Breakthrough news from Insilico Medicine! Their landmark AI-designed drug against lung disease showed promising results in Phase IIa trial! The company has just reported promising Phase IIa trial results for ISM001-055, an innovative drug candidate for idiopathic pulmonary fibrosis (IPF), developed entirely with generative AI. IPF affects nearly 5 million people worldwide, with a median survival of only 3-4 years after diagnosis. Current treatments can slow the disease, but none can stop or reverse it, highlighting the urgent need for new therapies. ISM001-055, designed with Insilico’s PharmaAI platform, specifically targets TNIK (Traf2- and NCK-interacting kinase) to address lung fibrosis at the molecular level. Trial Design and Results: ✔️ The double-blind, placebo-controlled Phase IIa trial involved 71 IPF patients across 21 sites in China, testing three dosing regimens (30 mg QD, 30 mg BID, and 60 mg QD) over 12 weeks. ✔️ Efficacy: Patients in the highest dose group showed a dose-dependent improvement in forced vital capacity (FVC), with a mean FVC increase of 98.4 mL, compared to a mean decline of -62.3 mL in the placebo group. ✔️ Quality of Life: The highest dose group saw meaningful improvement in Leicester Cough Questionnaire (LCQ) scores, indicating better quality of life and symptom relief. ✔️ Safety: ISM001-055 was well-tolerated across all doses, with only mild side effects such as gastrointestinal symptoms and liver function abnormalities. ✔️ US Trials: Insilico has begun recruiting for a Phase IIa trial in the United States, with aims to further validate ISM001-055’s efficacy and safety in a broader patient population. When I invested in the company back in 2016, I knew the ambitions of Alex Zhavoronkov would go a long way, but the project grew beyond even my expectations! They managed to build a real conveyor of drug candidates, PharmaAI. Since 2021, PharmaAI has helped nominate over 20 preclinical candidates across a portfolio of more than 30 assets, with 9 molecules receiving IND approvals. ISM001-055, discovered and designed using @PharmaAI’s Biology42 (PandaOmics) and Chemistry42 modules, is the latest example of how AI can make a transformative impact on complex, life-threatening diseases like IPF. Kudos to the team and hope for the patients! Image credit: Insilico Medicine

153

12 Comments

UCSF to Develop AI Monitoring Platform for Clinical Care Platform aims to bridge the gap between the rapid evolution of AI technologies and the need for robust, ongoing assessment of their efficacy, safety, and equity

1

Though benchmarks for medical AI are still imperfect, there's no question that AI will continue to march towards diagnostic excellence. These trends will place greater emphasis on physician-scientists – those who advance medical knowledge via closed loop clinic/lab interfaces, which can then feed back into large-scale AI systems that serve entire populations. This was all part of the OG vision for the learning health system. source: https://lnkd.in/eh2RM72s

101

11 Comments

The World's first non-invasive Bundle of His electrogram (bsBHE) - US patent Test time: 12 seconds Unit of measurement: millisecond AI mapping: 270 data (all invasive parameters) bsBHE is able to locate the block site (His bundle subbundle block) vs ECG: 1st degree AV block (unable to locate the block site) saahECG 1. saahECG enables screening of a larger number of patients before invasive procedures and allows dynamic monitoring after surgery (PTCA, RFA, PCI, pacemakers, bypass surgery). 2. saahECG has the ability to identify more hidden patients in need of a pacemaker. 3. saahECG is able to detect AVNRT/AVRT abnormalities even without any manifestation of disease. 4. Early identification of patients with latent disorders of the self-control system can be achieved by non-invasive methods. 5. saahECG can warn patients in advance of cardiac risk factors or the need for invasive treatment. 6. Many patients who need invasive tests often refuse them for various reasons. 7. Primary screening of patients in just 12 seconds becomes easy, fast and convenient. PhysioSign USA

27

1 Comment

Engaging Headline: Uniting Against Drug-Resistant Bacteria with AI Innovation World leaders have gathered to address a critical global issue: drug-resistant bacteria. With a staggering 4.95 million annual deaths linked to antimicrobial resistance, the urgency is palpable. Exploring the role of artificial intelligence, we are entering a new frontier in microbiology that could reshape the future of healthcare. Exciting advancements in AI are paving the way for breakthroughs in various areas: - Automation: Artificial intelligence can accelerate the drug discovery process by up to 50%, helping us combat new bacteria more effectively. - Identifying Antimicrobial Agents: AI helps researchers pinpoint promising new compounds at record speed, reducing development timelines from years to mere months. - Understanding Microbiomes and Proteomics: AI enhances our ability to analyze complex ecosystems and protein interactions, boosting the mapping of bacterial proteomes. By leveraging AI technology, we can envision a future where we effectively combat drug resistance and revolutionize our understanding of microbiomes. Together, we can and must act. We invite you to share your thoughts on the role of AI in fighting antimicrobial resistance. Join the conversation and follow Spin Quantum Tech for more insights! #Microbiology #ArtificialIntelligence #Laboratories #Future #Innovation #Health #QuantumComputing Spin Quantum Tech - SQT - Business acceleration through quantum computing. Dive into Quantum with Spin Quantum Tech: 📧: [email protected] 🏠: https://spinqtech.com 📱: +44 7537 188583

Dualism requires consciousness management We are still striving to find the code of the forefather with his abilities in order to make superhumans again, and at the same time solve a bunch of other problems. The existing methods for tracking the depths of ancestral lines available in the States only allow us to dive back 500 iterations, but this is clearly not enough to solve the basic problem. We are truly unique, we truly carry the experience of previous generations and we are the basis for the resurrection of all those who previously lived. The important thing is that it is precisely the small percentage that distinguishes us, but as a result of cutting granite we do not always get a perfect sculpture, which sees only this 1 percent of the required 100%. Taking this into account, there is no high probability that the resulting superhuman genome will be a good god... The rest will come from reflection.

1

Integrating AI into clinical trials is not just a trend—it's a game changer. The recent discussion with Gen Li, from Phesi highlights the transformative potential of AI to reduce costs, enhance privacy, and accelerate drug development. Here are some key takeaways from the interview: - Cost Reduction: AI-driven platforms can significantly cut the costs associated with clinical trials by streamlining processes, reducing manual efforts, and optimizing resource allocation. - Enhanced Privacy: With AI, data privacy is strengthened through advanced encryption methods and de-identification techniques, ensuring patient confidentiality. - Faster Drug Development: By automating data analysis and identifying patterns quickly, AI accelerates the timeline from research to market, benefiting patients sooner. - Increased Efficiency: AI tools help in better patient recruitment and retention by analyzing large datasets to find suitable candidates, thus increasing trial success rates. These innovations underscore the critical role AI will play in the future of clinical trials, making healthcare advancements more accessible and efficient. #ClinicalTrials #AIinHealthcare #HealthcareInnovation #DrugDevelopment #PrivacyInHealthcare

13

Google releases AlphaProteo: AI for Smarter Protein Design ! AlphaProteo is an AI system that designs new protein binders to target specific molecules - Unlike AlphaFold, which predicts how proteins interact, AlphaProteo actually creates proteins that can influence those interactions directly. Why AlphaProteo is a Game Changer: 🔬 Custom Protein Design: AlphaProteo can generate protein binders for tough targets like VEGF-A, linked to cancer and diabetes complications. 💡 Proven Results: 3 to 300x better binding affinity than current methods. Higher success rates across seven different target proteins, including SARS-CoV-2 and cancer-related proteins. 10x stronger binding for some proteins, like TrkA and BHRF1. 🚀 Real-World Impact: AlphaProteo has already shown it can prevent SARS-CoV-2 from infecting cells and binds more strongly to target proteins than existing methods—saving researchers time in the lab and opening new doors for breakthroughs. Check out their whitepaper here : https://lnkd.in/em-DSnem

44

2 Comments

Human Immunome Project unveils plan to map the human #immune system https://lnkd.in/gwPuHZWq Global research initiative aims to leverage immune monitoring technology and AI to decode the intricacy of the immune system. Today, on World Immunology Day, the global nonprofit scientific initiative Human Immunome Project (HIP) unveiled its scientific plan to map and model the human immune system. The immune system plays a pivotal role in human health, and the decline in its functionality as we age is linked to many age-related and chronic diseases. But, with less than 0.01% of the necessary data available to understand immune function and diversity globally, the complexity of the immune system has posed significant challenges to those seeking to understand it. The HIP is on a mission to decode the immune system, and claims that recent advancements in immune monitoring and artificial intelligence offer unprecedented opportunities to unravel its intricacies. “The immune system is the epicenter of human health, and our newly released scientific plan outlines an actionable, global strategy for how we can unlock the secrets of the immune system and harness its power to improve health for all,” said Dr Hans Keirstead, CEO of the Human Immunome Project. “The Human Immunome Project is at the cutting-edge of scientific discovery and is uniquely positioned to provide the scientific and medical communities with the knowledge and tools to innovate new diagnostics, drugs, vaccines and immunotherapies that can address disease, elongate healthspans, and enable more personalized and more effective healthcare for all,” added Keirstead, who is also Chairman of Immunis, a US biotech targeting age-related immune decline.

5

📢 FDA Digital Health Advisory Committee: Regulating Generative AI in Medical Devices. On Nov 20-21, the FDA’s Digital Health Advisory Committee will explore how to regulate Generative AI (GenAI)-enabled devices throughout the Total Product Lifecycle (TPLC). With 950 AI-enabled devices approved, all using "locked" algorithms at the moment, GenAI brings new challenges. Key focus areas: ✅ Premarket Evaluation – Foundational models need safety and effectiveness from the outset. ⚠️ Risk Management – Robust controls are essential for GenAI’s evolving nature, addressing data drift, bias, and unique risks like “hallucinations.” 🔍 Postmarket Monitoring – Ongoing oversight to ensure devices adapt safely in real-world use. 🔒 IEEE UL 2933: This standard supports the FDA’s priorities, providing a framework for trust, privacy, security, and safety across TPLC management. By promoting transparency and traceability, IEEE UL 2933 helps mitigate GenAI’s unique risks while aligning with FDA requirements to protect patient data and ensure device functionality—an essential foundation as GenAI begins to reshape healthcare technology.

9

Generative AI Lab introduces a new de-identification feature, enabling users to anonymize documents containing sensitive information such as PII (Personally Identifiable Information) and PHI (Protected Health Information). This functionality is intended to protect data privacy and ensure compliance with privacy regulations while preserving the data’s usefulness for subsequent analysis and processing. These projects allow the use of manually trained or pre-trained text-based NER models, together with prompts, rules, and custom labels created by the user for identifying sensitive data inside of tasks. Once the sensitive data is identified (either automatically or manually) and validated by human users, it can be exported for further processing. More information: https://lnkd.in/gMNvcrYB Generative AI Lab: https://lnkd.in/dik4i_hq #GenerativeAILab #AIUpdates #DeIdentification #DataPrivacy #PIIProtection #PHIProtection #AIinHealthcare #NLP #HealthcareAI #DataAnonymization #MachineLearning #AICompliance #TextAnnotation #Collaboration #AIEnhancements #QualityControl #DataSecurity #AIResearch #DataProcessing #HealthcareData #PrivacyRegulations #NLPTools

9

Read on for a recap of BIO 2024 San Diego 👇 1. AI and Digital Health  • Experimental validation in drug discovery via active feedback loops between silico and experimental data is crucial to ensure the integrity of AI findings, training algorithms, and adapting to and optimizing for safe and effective clinical outcomes. • Extended reality (VR, AR, and MR) combined with AI and biosensing, has a transformative impact on health issues such as anxiety, depression, and PTSD. (University of California, Berkeley)  • Overcoming systemic data bias via AI and digital health tools is leading to better access and lowering the overall cost of care, closing gaps in care delivery and diagnosis. (Cooley LLP) 2. Next Generation Biotherapeutics  • The longevity therapeutics field is growing in interest by investors in the diseases of aging framework and approaches, changing from preventative approaches to more pragmatic disease-based drug development and clinical trial designs. (The Wall Street Journal)  • Amid the opioid crisis, urgent innovation in pain therapeutics is vital - the NIH, small biotech, biopharma, and VCs can rejuvenate CNS & pain innovation to address this critical health need. (South Rampart Pharma, Inc.)  • Pharmaceutical innovations in space, such as microgravity, are being researched into how they may improve protein crystallization and reformulate existing pharmaceutical drugs. (Starburst Aerospace) 3. Oncology  • Radiopharmaceuticals can integrate concepts from personalized medicine. Collaboration between stakeholders is needed to drive future breakthroughs in radiopharmaceutical development and clinical trials. (BioCentury Inc.)  • 6 CAR-T cell therapies have been approved by the FDA over the last 6 years to treat blood cancers. Allogeneic cell therapy promises to be the next-generation option to broaden patient access, lower costs, and increase reliability and scalability. (Emily Whitehead Foundation)  • Co-opting biological processes is a key method for drug discovery, and a new generation of approaches include novel intra- and extracellular degraders, sequence-specific inhibitors of protein synthesis, and pH-engineered ADCs. (Venrock) 4. Cell and Gene Therapy (CGT)  • While CGTs are successful in the western world, manufacturing and supply chain are a major hurdle to reach patients in Asia, Middle East, Australia and Africa.  • Gene delivery vectors play a pivotal role in CGT by facilitating efficient delivery of genetic payloads to the cells. Novel delivery vehicles include next-generation AAVs, alternative viral vectors, and synthetic nanoparticles. (NorthX Biologics) Read the full recap of Biotechnology Innovation Organization 2024 here: https://lnkd.in/gSn7MtwG #Biointron #Antibodies #BIO2024 #Biotech #PharmaNews #DrugDevelopment #DrugDiscovery #Innovation #Networking #Education #Healthcare

8

🚀 Venture Bytes #108: Drug Discovery Process Major Beneficiary of AI Integration 🧪💊 In this edition, we dive into two exciting areas: 1️⃣ AI in Drug Discovery 🔬 The drug discovery process, known for its lengthy timelines, high costs, and low success rates, is being revolutionized by AI integration. AI can predict molecule properties and identify potential drug candidates, making AI-powered drug discovery startups attractive acquisition targets for contract research organizations (CROs). Notable startups include Generate:Biomedicines, BigHat Biosciences, Seismic Therapeutic, and BenchSci 💊 2️⃣ Data Labeling for AI Models 🏷️ Data labeling startups are becoming the gatekeepers of mass AI adoption, refining data for AI model utilization. As enterprise SaaS companies focus on integrating AI across their product suites, these startups are poised to achieve significant value in the coming years. Snorkel AI, WEKA and Scale AI are leading the charge. 📈 Plus, get our takes on: - Topical news items 📰 - Selected funding rounds 💰 - Exits 🎉 - Private and public valuations 📊 👀 The Liquidity Watchlist has one change: Gong.io has returned to the top-25 list, replacing WIZ. Our proprietary list features the top 25 promising TMT companies that could have a liquidity event in the next 12-18 months. Check out the full report 🔗: https://lnkd.in/essuuUv8 As always, Santosh Rao looks forward to your comments and suggestions. 💬 #VentureBytes #MarketResearch #InvestmentInsights #ArtificialIntelligence #AI #MachineLearning #ML #DrugDiscovery #BioTech #Pharma #CRO #DataLabeling #DataAnnotation #AIAdoption #SaaS #Enterprise #StartupFunding #Acquisitions #Exits #Valuations #LiquidityWatchlist #IPO #Merger #TMT #Technology #Media #Telecom #Investing #VentureCapital #PrivateEquity #Innovation #Disruption #FutureOfWork Manhattan Venture Partners

16

1 Comment

The world's first BioComputer, powered by lab-grown human 🧠 organoids, is now operational! As an ardent admirer of brain science and the scientific field at large, the progress in cultivating biocomputers is exciting to witness. However, when considering the ethical implications and cybersecurity perspectives, certain scientific breakthroughs appear too perilous to be realized. The potential risks they pose may significantly outweigh the advancements they promise, necessitating a cautious and responsible approach to their development and implementation. Thank you John Hagel for sharing. 🙏 #cybersecurity #cyberethics #biocomputer #biocybernetics #braincomputer #computing #bioengineering

5

⚕ Harvard Med researchers recently trained & tested a new large vision model (LVM) called CHIEF for cancer detection. Results: Overall accuracy of ~94% for cancer detection, rising to 96% cent for oesophagus, stomach, colon & prostate tumors with a new capability to link tumor cell patterns to genomic mutations and related cell aberrations to recommend treatments without the need for costly/ slow DNA sequencing. I can't tell you how many people I know now who need this. Expect to see more of this. AI will transform healthcare outcomes. Not today or tomorrow or even in 3-5 yrs. But in 5-10 yrs absolutely. It will be like we were living in the dark ages today. Keep in mind today is the worst that the models will perform. That said, it will take time to collect, clean, and apply the necessary massive data sets. 👇 CHIEF was trained on 15m unlabelled sections of images & then 60k images of tissues, covering 19 different cancers. Performance was evaluated on ~20k images from 24 hospitals & patient cohorts around the world. 👀 An incredible feat that will have almost unlimited/ priceless benefits.

5

2 Comments