Weekly Tech+Bio Highlights #76: Neurons, Genomes, and Self-Driving Labs
Lilly uses digital twin AI to scale GLP-1 production, generative genomics—now in Nature, $230M for a rice-grain-sized retinal implant
This week was rather AI—Insilico’s AI-designed anemia drug entered Phase I trials, Ginkgo opened up cloud access to its robotic labs where you can submit experiments in plain English, Eli Lilly revealed it's using AI-powered digital twins (of its factory) to optimize GLP-1 production, and another AI diagnostic system got FDA clearance for stroke detection. Agentic AI keeps creeping into everything, and the line between "AI-assisted" and "AI-driven" is getting blurrier.
Foundational and open source: Arc Institute published its Evo 2, a 40B-parameter genomic foundation model that can generate and analyze DNA sequences, in Nature, while French Raidium released an open-source foundation model for reading CT and MRI scans.
Meanwhile, a Neuralink co-founder pulled in $230M for a rice-grain-sized retinal implant, researchers uploaded a full fruit fly brain connectome into a simulated body and got it to move, and Japan approved the first iPSC-derived therapies for heart failure and Parkinson's.
Hi! This is BiopharmaTrend’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:
🤖 AI x Bio
(AI applications in drug discovery, biotech, and healthcare)
🔹 AI boosts drug manufacturing — Eli Lilly, led by CIO Diogo Rau, used AI-powered digital twins of its factories to optimize production and defect detection for its blockbuster GLP-1 drugs, enabling significantly higher output of its diabetes and weight-loss treatments.
🔹 Open-source radiology foundation model — Raidium released Curia-2, an open-source AI model for analyzing CT and MRI scans that detects ~200 radiological findings and achieved 88.6% accuracy on 3D imaging benchmarks, with a 2B-parameter version available on Hugging Face.
🔹 Generative AI for DNA design — Arc Institute, Stanford University, and NVIDIA published Evo 2 in Nature, a 40B-parameter open-source genomic foundation model that can generate and analyze DNA sequences, with experiments showing AI-designed sequences can alter chromatin accessibility in living cells.
🔹 Automating protein lead optimization — Dutch company Cradle released the CRADLE-1 framework on bioRxiv, a machine-learning system that automates protein lead optimization across modalities (antibodies, enzymes, peptides, vaccines, gene-editing systems), achieving ~90-95% success in meeting target product profiles in validated campaigns.
🔹 On-premise AI for drug discovery — Liquid AI and Insilico Medicine released a 2.6B-parameter foundation model designed to run on private pharmaceutical infrastructure, enabling tasks like molecular design, ADMET prediction, and retrosynthesis while keeping sensitive data on-premise.
🔹 AI-designed anemia drug enters clinic — Insilico Medicine announced that its AI-discovered oral therapy for chronic kidney disease–related anemia has begun Phase I trials after first patient dosing by TaiGen Biotechnology, following a licensing deal for Greater China rights in December 2025.
🔹 Cloud access to robotic labs — Ginkgo Bioworks launched Ginkgo Cloud Lab, a web-based platform where researchers can submit natural-language experiment protocols that AI agents translate into automated workflows executed on robotic lab infrastructure with 70+ instruments.
🔹 AI-enabled nanobody discovery — MindWalk Holdings launched the B Cell Llama platform, combining llama-derived VHH nanobody discovery with AI-guided candidate selection to develop bispecific antibodies and cell therapies, supported by a peer-reviewed study showing enhanced potency and cross-variant activity.
🔹 AI-driven small-molecule discovery — Ono Pharmaceutical expanded its collaboration with Congruence Therapeutics to discover small-molecule drugs for neurology and immunology using AI models that analyze dynamic protein structures and identify hidden binding pockets.
🔹 Dynamic AI models for biology — Sentinal4D, led by Chris Bakal, is developing AI systems trained on time-resolved 3D imaging of living cells to model cell behavior and drug responses, addressing limitations highlighted by Eli Lilly CEO Dave Ricks and AstraZeneca bioinformatics director Ming Tang that current AI relies mostly on static biological data.
🔹 AI smart goggles for labs — Stanford University and Princeton University researchers developed LabOS, an AI-powered smart-goggle system using NVIDIA vision-language models that watches scientists’ hands during experiments and guides protocols in real time to prevent mistakes and help novices perform like experts.
🔹 AI “virtual pharma” study — Stanford University researchers simulated a virtual drug company using ~37,000 AI agents to analyze 56,000 clinical trials, finding that drugs targeting cell-type-specific genes are 48% more likely to reach market and show 32% fewer adverse events.
🔹 Self-driving lab discovers new lipid designs — Researchers in a Cell study developed the LUMI-lab autonomous system that combines robotics and AI to explore 221,000 lipid candidates for mRNA delivery, discovering brominated lipid structures that improved delivery efficiency and achieved ~20% in vivo CRISPR editing in lung cells, with the platform’s code and hardware released open source.
🔹 FDA-cleared AI for early stroke detection — Harrison.ai received FDA clearance for an AI system that analyzes routine non-contrast brain CT scans to flag early signs of ischemic stroke and prioritize urgent cases for faster clinical review. Also: Harrison.ai expanded its imaging AI platform by adding AIRAmed, Koios Medical, Lunit, and Nanox AI, enabling hospitals to access multiple AI tools for X-ray, CT, MRI, mammography, and ultrasound through a single vendor-neutral infrastructure.
🔹 Amazon Web Services launched Amazon Connect Health, a suite of AI agents that automate tasks like patient identity verification, appointment scheduling, clinical note generation, medical coding, and summarizing patient histories for clinicians.
🔹 Generalist medical AI model — Harvard researcher Pranav Rajpurkar and collaborators introduced MedVersa in NEJM AI, a multimodal foundation model that analyzes medical images and clinical text to generate diagnoses and reports, outperforming several specialized systems and reducing radiology reporting errors and time in clinical tests.
🔹 AI for rare disease diagnosis — Nature study reports that the DeepRare agentic AI system integrates 40+ medical and genomic analysis tools to diagnose rare diseases with traceable reasoning, reportedly outperforming experienced physicians and attracting registrations from 500+ medical institutions since July 2025.
🔹 AI agents for bioinformatics infrastructure — recently launched Phylo introduced Biomni, an AI research agent that lets scientists run large-scale bioinformatics pipelines and models like AlphaFold using natural language while orchestrating GPU-heavy HPC workloads, massive datasets, and asynchronous jobs in specialized research environments.
AI, Generally:
🔹 Autonomous AI research loop — Andrej Karpathy open-sourced an AI agent that autonomously modifies training code and runs ~100 machine-learning experiments overnight on a single GPU using a fixed 5-minute evaluation loop, demonstrating how agentic systems can iteratively improve models with minimal human input through a simple Markdown-defined research strategy.
🔹 Limits of current AI reasoning — On the a16z podcast, Vishal Misra and Martin Casado discuss how modern deep learning systems optimize for Shannon entropy (statistical pattern prediction) but lack Kolmogorov-style causal mechanism discovery, arguing that current AI cannot generate Einstein-level scientific theories because it cannot extend its hypothesis space or perform counterfactual causal reasoning.
🔹 AI discoveries humans may not understand — Asimov Press highlights the emerging “legibility problem,” where increasingly autonomous AI systems could generate scientific discoveries beyond human comprehension, raising concerns about whether humans will be able to interpret, validate, or apply AI-generated knowledge.
💰 Money Flows
(Funding rounds, IPOs, and M&A for startups and smaller companies)
🔹 Retinal implant for vision restoration — Science Corporation, founded by Max Hodak (Neuralink co-founder), raised $230M Series C (total $490M) to commercialize a rice-grain–sized retinal implant that converts infrared signals from smart glasses into electrical stimulation, restoring partial vision in patients with degenerative eye diseases.
🔹 AI-driven microbiome modeling —Outpost Bio raised $3.5M pre-seed to build a “lab-in-the-loop” platform combining automated microbiology experiments with machine learning to model and predict changes in the human microbiome.
🚜 Market Movers
(News from established pharma and tech giants)
🔹 Tempus AI expanded a multi-year collaboration with Merck to use multimodal clinical and molecular datasets with AI analytics to identify oncology biomarkers, study treatment resistance, and guide precision medicine drug development.
⚙️ Other Tech
(Innovations across quantum computing, BCIs, gene editing, and more)
🔹 Living neurons play video games — Cortical Labs demonstrated that ~200,000 lab-grown human neurons connected to microelectrode arrays can learn to play the video game Doom by translating game visuals into electrical stimulation and using neural activity to control in-game actions, highlighting early progress toward biological computing systems.
🔹 Embodied whole-brain emulation — The Innermost Loop reports how Eon Systems demonstrated a connectome-derived digital model of the entire fruit fly brain (~125k neurons, ~50M synapses) controlling a physics-simulated body to generate multiple behaviors, marking a milestone toward scalable brain emulation and future mouse- and human-scale digital brains.
🔹 First iPSC therapies approved — Cuorips and Sumitomo Pharma received Japan’s first approval for induced pluripotent stem cell–derived treatments targeting heart failure and Parkinson’s disease, marking the first commercialization of iPSC regenerative therapies.
🏛️ Bioeconomy & Society
(News on centers, regulatory updates, and broader biotech ecosystem developments)
🔹 Leadership change at FDA biologics division —Vinay Prasad, director of the FDA’s Center for Biologics Evaluation and Research, will leave the agency at the end of April after a contentious tenure overseeing vaccines and gene therapies, with FDA Commissioner Marty Makary confirming a search for his successor.
Read also:
Three Big Ideas in Aging Research That Could Shift the Therapeutic Landscape