Weekly Tech+Bio Highlights #12
ALSO: A $1.9B Health Data Tech Gamble in Question; Novel 3D-printed Lung Model; Novel AI Framework for Detecting LLM "Hallucinations" in Medical Summaries; Arguably the Largest ‘AI in Bio’ Deal.
Hi! I am Andrii Buvailo, and this is my weekly newsletter, ‘Where Tech Meets Bio,’ where I talk about technologies, breakthroughs, and great companies moving the biopharma and medtech industries forward.
If you've received it, then you either subscribed or someone forwarded it to you. If the latter is the case, subscribe by pressing this button:
Now, let’s get to this week’s topics!
Quick News Highlights
💰 Genentech has partnered with Sangamo to develop an epigenetic Alzheimer’s therapy targeting tau protein, in a deal worth up to $1.95 billion. Sangamo, facing financial struggles, will receive $50 million upfront and shift focus to neuroscience, leveraging its new AAV capsid for better brain delivery.
🔬 Neuralink has successfully implanted its brain-computer interface in a second patient, allowing those with spinal cord injuries to control digital devices through thought alone. Elon Musk expects eight more patients to receive implants this year as part of ongoing clinical trials.
Watch my interview with Dr. Brian Jamieson, CTO of Diagnostic Biochips, were we talk about the opportunities and challenges of brain computer interfaces (BCI) technology, Neuralink, and the transformative potential of BCIs in neuroscience and pharmaceutical research:
Ex-NASA Expert Unveils Everything You Need to Know About Brain-Computer Interfaces
🌍 Alán Aspuru-Guzik, Professor of Chemistry and Computer Science at the University of Toronto, co-founder of Zapata AI, Kebotix, and Intrepid Labs, and a pioneer in quantum computing and AI for materials discovery, is co-organizing the Generative AI for Environmental Sustainability Consortium, together with Insilico Medicine, an AI-first drug discovery company.
This initiative focuses on developing AI models for CO2 capture, hydrogen storage, novel fuels, lubricants, and agriculture.
💰 Schrödinger secured a $10M grant from the Gates Foundation to expand its computational platform, using AI and physics-based methods to predict toxicity risks in drug development. The project aims to model off-target protein binding, potentially reducing failures and benefiting global drug safety.
🤝 Seqera has acquired tinybio, a NYC-based tech-bio startup specializing in AI-integrated bioinformatics tools, to enhance data analysis accessibility for scientists. This partnership aims to democratize bioinformatics through GenAI and large language models (LLMs).
🔬 Harvard's Wyss Institute and SEAS researchers have advanced 3D bioprinting by developing a method to print complex, multilayered blood vessels, bringing the possibility of implantable human organs closer to reality.
🏥 Tampa General Hospital implemented Apella, an advanced AI platform leveraging predictive analytics, computer vision, and real-time data to optimize operating room efficiency.
🔬 Researchers at the University of Oxford developed a powerful protein-based biological age clock using data from over 51,000 participants, accurately predicting aging, disease risk, and premature death across diverse populations by analyzing 204 blood proteins.
🔬 Insilico Medicine's AI-designed TEAD inhibitor, ISM6331, received FDA IND approval and Orphan Drug Designation for mesothelioma, with plans to initiate U.S. clinical trials.
🔬 UCLA Health Jonsson Comprehensive Cancer Center launches a first-in-human clinical trial for a personalized cancer vaccine targeting H3 G34-mutant diffuse hemispheric glioma in adolescents and young adults, funded by the Department of Defense, aiming to improve survival rates and provide insights into immune responses to brain cancers.
💰 Riverlane secured $75M in Series C funding to advance its Deltaflow quantum error correction technology, which integrates QEC chips and software to correct billions of qubit errors per second.
💰 Biomatter has raised €6.5M in seed funding to advance its Intelligent Architecture™ platform, which uses generative AI to design novel enzymes from scratch. This platform aims to targets industries like biomanufacturing, agriculture, and life sciences by reducing the time needed to create custom enzymes, already being used by partners like Thermo Fisher and BASF.
A $1.9B Health Data Tech Gamble in Question
The news article discusses Roche's consideration of divesting or restructuring Flatiron Health, a cancer data specialist company that Roche acquired for $1.9 billion in 2018.
Flatiron Health, based in New York and founded by former Google executives, manages electronic patient records for a large network of cancer clinics in the U.S. The company mines this data and sells it to pharmaceutical companies, which use it to support their research and development (R&D) efforts.
What is Happening:
Roche, a leading global developer of cancer drugs, purchased Flatiron Health as part of a broader strategy to integrate health technology into its operations, especially in oncology. Flatiron's data resources were expected to enhance Roche's R&D capabilities.
Since Roche acquired Flatiron Health, the company has faced several challenges:
Impact on Business Relationships: Roche's ownership of Flatiron has discouraged some rival pharmaceutical companies from doing business with the start-up. These competitors may be wary of sharing sensitive data with a company owned by a major player in the same market.
Sales Decline: As a result of these strained relationships, Flatiron's sales have suffered. The company generates about two-thirds of its revenue from selling data to pharmaceutical companies, so a reduction in clients has had a significant impact.
Leadership Changes: Key Roche executives who originally supported the acquisition have since left the company, leading to a decrease in internal advocacy for Flatiron within Roche
Roche's Strategic Review: In response to these challenges, Roche has enlisted Citigroup to explore options for Flatiron Health, which could include selling the company or part of it, or possibly bringing in a partner to help manage the business. However, the review may not necessarily result in the sale of the company.
Novel 3D-printed Lung Model
Researchers from POSTECH and KRICT have developed a 3D bioprinted lung model that could significantly improve research on respiratory diseases like COVID-19.
Unlike traditional 2D cell cultures or animal models, this artificial lung closely mimics the structure and function of a human lung, including key features that make it susceptible to viral infections. The model can maintain its structure for 21 days, allowing scientists to observe how the virus affects lung cells over time, which is much longer than traditional models that only last a few days.
This extended time frame enables researchers to study how the virus spreads and interacts with the immune system more accurately. The team also used this model to test how well COVID-19 drugs, like Remdesivir and Molnupiravir, work by observing how these drugs move through the lung tissue and inhibit the virus.
This approach provides a more realistic assessment of drug effectiveness and potential side effects, which could help speed up the development of new treatments. The research emphasizes the potential of using advanced 3D bioprinting technology to improve our preparedness for future respiratory virus outbreaks.
Novel AI Framework for Detecting LLM "Hallucinations" in Medical Summaries
Mendel AI, a startup specializing in healthcare artificial intelligence (AI), in collaboration with the University of Massachusetts Amherst (UMass Amherst), has unveiled new research focused on addressing the challenge of "faithfulness hallucinations" in AI-generated medical summaries.
The research centers on large language models (LLMs) like GPT-4o and Llama-3, which have shown potential in generating medical summaries. However, these models are prone to hallucinations—instances where the AI generates incorrect or misleading information. These inaccuracies pose significant risks in medical contexts, potentially leading to misdiagnoses or inappropriate treatments.
The study categorizes hallucinations into five distinct types and introduces a detection framework designed to identify these errors systematically. A pilot study involving 100 medical summaries generated by GPT-4o and Llama-3 revealed that while GPT-4o produced longer summaries, it often made erroneous reasoning leaps, leading to more hallucinations. In contrast, Llama-3 generated fewer hallucinations by avoiding extensive inferences but at the cost of summary quality.
The detection framework identified specific inconsistencies in the models, including medical event inconsistencies, incorrect reasoning, and chronological errors. For instance, GPT-4o was found to have higher instances of incorrect reasoning and inconsistencies, whereas Llama-3 had fewer errors but produced lower-quality summaries.
The Hypercube System: A Tool for Automated Detection of LLM "Hallucinations"
To tackle the issue of hallucinations, the research explored automated methods that could mitigate the high costs and time associated with manual review. Central to this effort is the Hypercube system, which integrates medical knowledge bases, symbolic reasoning, and natural language processing (NLP) to detect hallucinations. This system provides a comprehensive representation of patient documents, allowing for an initial automated detection step before human expert review.
The ongoing integration of AI into healthcare makes addressing hallucinations in LLM outputs increasingly crucial. Future research will focus on refining detection frameworks and exploring more advanced automated systems like Hypercube. The goal is to achieve the highest levels of accuracy and reliability in AI-generated medical content.
The academic community has recognized the significance of this work. The research paper titled "Faithfulness Hallucination Detection in Healthcare AI" has been accepted for presentation at the KDD AI conference in August 2024, detailing the methodologies and technologies behind Hypercube’s success.
In Arguably the Largest ‘AI in Bio’ Deal, Recursion Acquires UK-based "End-to-End" AI Biotech Exscientia
Recursion Pharmaceuticals, a biotech firm leveraging artificial intelligence (AI) for drug discovery, has announced its acquisition of Exscientia, a smaller rival, for $688 million in an all-stock transaction. The deal, expected to be publicized on Thursday morning, underscores the growing reliance on AI within the pharmaceutical industry to expedite drug development and reduce associated costs.
Human clinical trials represent the most costly and time-consuming phase of drug development, often requiring several years and over a billion dollars to bring a new drug to market. By integrating AI, companies aim to streamline patient recruitment for trials and minimize the number of participants needed, thereby accelerating development timelines and saving substantial financial resources.
Founded in 2013, Recursion is a phase 2 clinical-stage company with a focus on treatments for rare diseases and certain cancers. The firm went public in 2021 and maintains collaborations with major pharmaceutical entities such as Roche AG and Bayer. Earlier this year, Recursion unveiled BioHive-2, a supercomputer powered by Nvidia technology, designed to accelerate drug discovery processes. Nvidia is also an investor in Recursion.
Read also: A Landscape of AI-discovered Molecules and Target Novelty Analysis
Exscientia, headquartered in the UK, operates an AI-driven drug discovery platform and is developing a pipeline targeting immunology and oncology. The acquisition of Exscientia is anticipated to enhance Recursion's drug development pipeline and extend its partnerships with leading pharmaceutical companies, including Sanofi and Merck KGaA.
Under the terms of the agreement, Exscientia shareholders will receive 0.7729 Recursion shares for each Exscientia share held. The transaction is expected to close in early 2025, providing the merged entity with approximately $850 million in cash, sufficient to support operations for the next three years.
Salt Lake City-based Recursion will report its quarterly earnings later on Thursday. Advisors Allen & Co., Wilson Sonsini Goodrich & Rosati, Clifford Chance for Recursion, Centerview Partners, and A&O Shearman for Exscientia facilitated the acquisition.
AI in Drug Discovery Sector Consolidation
Recursion Pharmaceuticals has been actively expanding its capabilities in AI-driven drug discovery through previous acquisitions. The acquisition of Valence Discovery in 2023 brought in advanced machine learning models that predict molecular properties and interactions, aimed at enhancing Recursion's ability to identify promising drug candidates with precision.
Additionally, the acquisition of Cyclica in 2023, introduced Ligand Express and Ligand Design platforms, which utilize AI to predict drug-target interactions and design novel molecules, enriching Recursion's computational chemistry capabilities. These strategic moves have integrated sophisticated computational methods into Recursion's biology-centric platform, optimizing their drug discovery and development processes.
In its turn, Exscientia previously strengthened its AI capabilities by acquiring Allcyte back in 2021. Allcyte's focus on AI-powered precision medicine for hematologic cancers employs high-content imaging and AI to analyze patient-derived tumor samples, offering insights into drug efficacy and enabling personalized treatment strategies. This acquisition allowed Exscientia to merge functional drug testing with AI-driven design, advancing targeted oncology therapies.
The recent acquisition of Exscientia by Recursion for $688 million is a strategic move to consolidate Recursion's position as a leading player in the AI-driven biopharmaceutical sector. By integrating Exscientia’s AI platforms and its drug pipelines focused on immunology and oncology with Recursion's extensive dataset and machine learning infrastructure, including the Nvidia-powered BioHive-2 supercomputer, Recursion aims to create a comprehensive end-to-end drug discovery and development process. This deal signifies Recursion's ambition to leverage combined technological advancements to accelerate the discovery and delivery of innovative therapies, solidifying its leadership in the AI space
This is reflective of a larger pattern of AI in drug discovery sector consolidation, which is reflected in a dozens of acquisition deals over the recent years:
