Weekly Tech+Bio Highlights #32: Chemical Lab Run by Multi-Agent AI
AI in Materials Science, Humanoid Lab Robots, Summarizing Findings From 22M Scientific Papers, and more...
Hi! This is
’s weekly newsletter, Where Tech Meets Bio, where we explore technologies, breakthroughs, and cutting-edge companies.If this newsletter is in your inbox, it’s because you subscribed or someone thought you might enjoy it. In either case, you can subscribe directly by clicking this button:
Let’s get to this week’s topics!
Multi-Agent AI Runs the Lab
Researchers from the University of Science and Technology of China (USTC), along with a collaborator from the University of Birmingham in the UK, have developed a multi-agent robotic AI chemist designed to automate complex chemical research with minimal human intervention. The system, called ChemAgents, combines large language models (LLMs) and laboratory robotics to independently conduct multi-step experiments, and is designed to accelerate discovery and optimization of new materials and chemical processes.
According to the study, ChemAgents is built around Llama-3-70B, an LLM integrated into a hierarchical multi-agent system. It consists of five core agents: a Task Manager that oversees operations and four specialized agents—Literature Reader, Experiment Designer, Computation Performer, and Robot Operator—each tasked with different stages of the research process. The system employs four key resources: a scientific literature database, a protocol library for experimental design, a machine learning (ML) model library, and an automated lab with robotic infrastructure.

Literature Reader searches a local database of around 1.2 million scientific publications and extracts relevant information to inform experimental design.
Experiment Designer translates this information into detailed experimental protocols by searching a library of templates and adjusting them as needed for specific tasks. The protocol library includes templates for over 20 experimental stations that cover a range of chemical processes (liquid dispensing, magnetic stirring, infrared spectroscopy, X-ray diffraction, photocatalysis, electrocatalysis, etc.). If no matching template is found, the Experiment Designer can query the available stations and design a new protocol from scratch.
Robot Operator converts these protocols into executable robot commands, coordinating movements and manipulations within the automated lab. The lab includes a mobile robot for navigating the workspace and a benchtop robotic arm for conducting operations such as dispensing, stirring, and spectroscopic analysis.
For data analysis and optimization, the Computation Performer accesses pre-trained ML models from the model library. It can adjust these models using experimental results and incorporates a Bayesian optimizer for refining experimental conditions through iterative feedback.
The researchers demonstrated ChemAgents’ capabilities through six experimental tasks—from basic synthesis and characterization to complex parameter screening and functional material discovery. According to the authors, it successfully executed these tasks with minimal human input, suggesting that the system could enable faster, more efficient chemical research.
One example involved ChemAgents discovering and optimizing a high-entropy catalyst for the oxygen evolution reaction, reportedly achieving improved catalytic performance through literature mining, automated synthesis, and iterative refinement.
Core idea highlighted in the study is the hierarchical design that should allow agents to collaborate and share outputs. This approach aims to increase flexibility and scalability compared to single-agent systems. The authors suggest that ChemAgents could reduce the time and cost of chemical research while improving reproducibility and enabling non-experts to engage with complex experimental workflows.
AI in Materials Science
Scientists are increasingly turning to artificial intelligence to design new materials for applications ranging from carbon capture to advanced semiconductors. According to a recent article in The Economist, AI models are being used to identify and optimize complex materials like metal-organic frameworks (MOFs), which have the potential to improve carbon capture, battery performance, and bioplastics.
AI's ability to accelerate discovery is already being measured—study by Aidan Toner-Rodgers (MIT) analyzed the effects of an AI tool on the productivity of materials researchers at a large American company. The study found that AI increased the number of materials discovered by 44%, boosted the number of prototypes using those materials by 17%, and raised the number of patents filed by 39%.
Stef van Grieken, co-founder of Cradle, an AI protein lab, compared AI-driven materials discovery to drug development. He described the pharmaceutical industry as “private equity with laboratories attached,” explaining that clinical trials distribute risk and reward throughout the industry, allowing for greater investment in early-stage research. Materials science, by contrast, lacks a comparable financial structure, making it harder for companies to commercialize new materials, as they often require custom manufacturing facilities and significant capital investment to reach commercialization, which creates a bottleneck even when AI accelerates discovery.
Our interview with Stef van Grieken on AI-Driven Protein Engineering.
It seems automation is beginning to address this gap. There are examples of AI-enabled lab systems for designing, synthesizing, and testing materials with minimal human input. However, as van Grieken noted, materials production remains complex—minor differences in humidity or air quality can derail the synthesis process, complicating the path from discovery to practical application. The real test for AI in materials science will be whether it can overcome these scaling challenges and help bridge the gap between research and production.
This newsletter reaches over 8.1K industry professionals from leading organizations across the globe.
Interested in sponsoring?
Contact us at info@biopharmatrend.com
🤖 AI x Bio
(AI applications in drug discovery, biotech, and healthcare)
🔹 Elsevier launches ScienceDirect AI to streamline research—The new tool uses generative AI to summarize findings from 22 million scientific papers, helping researchers cut literature review time by up to 50% while ensuring traceability and accuracy.
🔹 Insilico raises $110M and debuts humanoid lab robot—The AI-driven drug discovery company secured new funding to support clinical trials and introduced "Supervisor," a robot designed to mimic scientists’ lab techniques and enhance automation.
🔹 Nosis Biosciences and Daiichi Sankyo partner to develop AI-driven RNA therapies targeting hard-to-reach organs like the heart, brain, and lungs, expanding RNA treatment potential beyond the liver. Nosis also partnered with Johnson & Johnson.
🔹 AI-designed protein mimics 500 million years of evolution—Researchers developed ESM3, a protein language model that generates functional proteins based on user prompts, creating a novel fluorescent protein with only 58% similarity to known variants.
🚜 Market Movers
(News from established pharma and tech giants)
🔹 Tempus expands clinical trial reach with AI-powered Deep 6 acquisition—The deal adds over 750 provider sites and 30M+ patients to Tempus' network, using AI to match patients to trials and cut recruitment time.
🔹 BMS snaps up 2seventy bio for $286M—Bristol Myers Squibb is acquiring longtime partner 2seventy bio, gaining full rights to their jointly developed CAR-T therapy Abecma for multiple myeloma.
🔹 Roche dives deeper into weight loss drugs with $1.65B deal—Roche is paying Zealand $1.65B upfront (plus up to $3.6B in milestones) to co-develop and commercialize an amylin analog for obesity, in one of the largest upfront payments for a single obesity asset. Also: Stanford used machine learning to identify a peptide that reduces appetite and weight without Ozempic’s side effects (in mice, pigs).
💰 Money Flows
(Funding rounds, IPOs, and M&A for startups and smaller companies)
🔹 UK biotech raises £600K to advance 3D lung model as an alternative to animal testing—ImmuONE’s ImmuLUNG platform aims to improve drug and chemical safety testing, addressing the 90% failure rate of drugs that succeed in animals but fail in humans.
🔹 Entos Pharmaceuticals secures $198.5M to build biomanufacturing facility—Backed by $77.5M from Canadian and Alberta governments, Entos will develop next-gen genetic therapies using its Fusogenix platform, expanding Canada's role in drug manufacturing.
🔹 Hinge Health’s AI-powered physical therapy platform files for IPO—Hinge’s wearable sensor and AI-driven motion tracking tech helped the digital therapy company generate $390M in 2024, with plans to raise $500M in its NYSE debut.
🔹 Illumina braces for China sales loss with $100M cuts—After China banned Illumina’s DNA sequencers in retaliation for new U.S. tariffs, the company plans to cut $100M in expenses and shift focus to other international markets while targeting high-single-digit revenue growth by 2027.
⚙️ Other Tech
(Innovations across quantum computing, BCIs, gene editing, and more)
🔹 Beam’s gene editing therapy rewrites DNA to treat lung and liver disease—Beam Therapeutics’ precision base editing tech boosted protective protein levels and reduced harmful ones in a Phase 1/2 trial for alpha-1 antitrypsin deficiency, with no serious side effects reported.
🔹 Boehringer Ingelheim and Veeva launch new R&D platform—built on Veeva’s development cloud, the platform aims to streamline clinical, regulatory, and quality data processes to accelerate drug development and improve trial efficiency.
🔹 Siemens Healthineers expands photon-counting computed tomography lineup—FDA clears two new models, including its first single-source system for broader use in emergency rooms, enhancing speed and precision with AI-driven imaging and faster scan times.
🔹 Helius Medical Technologies launches new AI-powered brain-computer interface (BCI) subsidiary—Revelation Neuro, focused on non-implantable BCI tech for motor function rehab, will leverage Helius' existing $70M neuromodulation investment and clinical data from 400+ subjects, aiming for personalized neurorehabilitation.
🏛️ Bioeconomy & Society
(News on centers, regulatory updates, and broader biotech ecosystem developments)
🔹 Advanced biotech could cut global emissions by 5% and add $1T in annual value, according to a report from the Advanced Biotech for Sustainability (AB4S) coalition, which highlights the need for $500B in investment by 2040 to scale bio-based solutions across industries like food, pharma, and construction.
🔹 NIH grants fueled $94B in economic activity and over 400K jobs in 2024, according to a report from United for Medical Research, which found that every $1 of NIH funding generated $2.56 in economic returns, up from $2.46 in 2023.
🔹 Despite sweeping layoffs across federal agencies, including the IRS and USAID, the FDA’s drug and device review departments have so far remained unaffected.
🚀 New Kids on the Block
(Emerging startups with a focus on technology)
🔹 Élancé Therapeutics, a new biotech spinout from China’s Harbour BioMed, will focus on developing bispecific antibodies aimed at reducing body weight while preserving muscle mass. The company will leverage Harbour's antibody technology and its “Hu-mAtrIx” AI platform from its Nona Biosciences subsidiary to design more effective therapies. Hu-mAtrIx uses machine learning to analyze complex biological data, helping to identify optimal bispecific antibody targets and predict their efficacy in modulating multiple metabolic pathways.
🔹 MeiraGTx and Hologen launch AI-driven gene therapy startup with $200M boost—Hologen Neuro AI will develop gene therapies for Parkinson’s and obesity using generative AI to improve drug discovery and manufacturing. MeiraGTx will lead clinical development and manufacturing, while Hologen’s AI will refine manufacturing efficiency and enhance trial design. This deal builds on growing interest in AI-driven drug development, with Hologen’s team including former Google CEO Eric Schmidt and QuantHouse co-founder Pierre-François Filet.