Welcome to the second PythRaSh's AI Newsletter of 2026! This week marks a historic moment as Eli Lilly and NVIDIA announced a $1 billion AI co-innovation lab at the J.P. Morgan Healthcare Conference. Arc Institute released Evo 2, the largest publicly available genomic foundation model trained on 9.3 trillion nucleotides, achieving 90% accuracy in predicting disease-causing mutations. UC Berkeley and UCSF's Pillar-0 medical imaging AI outperformed all public competitors while being 150x faster.
Yet as breakthroughs accelerate, Pacific AI documents "The End of Voluntary Ethics," with global shift to enforceable AI law creating a "compliance splinternet." This week crystallizes healthcare AI's dual reality: extraordinary technical progress amid escalating regulatory fragmentation.
🚀 EVENT OF THE WEEK
At the 2026 J.P. Morgan Healthcare Conference (January 12-15), pharmaceutical giant Eli Lilly and AI chipmaker NVIDIA announced they will establish a new AI co-innovation lab in the San Francisco Bay area, with companies jointly investing up to $1 billion over five years to advance drug discovery.
Lab Structure and Mission:
The lab will bring together Lilly's biology and scientific experts with NVIDIA's AI model builders and engineers in a shared facility, with initial focus on generating large-scale biological and chemical datasets and developing AI models to shorten drug development timelines. Operations begin early 2026.
Why This Matters:
This co-location model creates integrated teams where biologists, chemists, and AI engineers work together from project inception, enabling iterative cycles of computational prediction, experimental validation, and model refinement—the "closed-loop discovery" that pharmaceutical leaders have identified as the future. The $1 billion scale signals industry confidence that AI will fundamentally transform drug discovery, not merely accelerate existing methods.
Key Implications: $1 billion investment marks largest pharma-AI partnership to date. NVIDIA positioning as end-to-end pharmaceutical AI infrastructure provider. May accelerate industry consolidation around large pharma-tech partnerships. Raises questions about access, affordability, and data governance in AI drug discovery.
⚡ QUICK UPDATES
- 🧬 Evo 2: Arc Institute Releases Largest Public Genomic Foundation Model: Arc Institute developed Evo 2, trained with 40 billion parameters on over 9.3 trillion nucleotides from 100,000+ species. Achieves 90% accuracy predicting disease-causing BRCA1 mutations and can design genomes up to 1 megabase in length. Novel StripedHyena 2 architecture enables training 3x faster than transformers. Publicly accessible via Arc's GitHub and NVIDIA BioNeMo. Read More
- 🖼️ UC Berkeley & UCSF Release Pillar-0: Top-Performing Open-Source Medical Imaging AI: Pillar-0 interprets 3D volumes directly, recognizing hundreds of conditions from single CT or MRI exams. Achieved .87 AUC across 350+ findings, outperforming Google's MedGemma, Microsoft's MI2, and Alibaba's Lingshu. Atlas architecture 150x faster than traditional transformers. Complete codebase publicly released. Read More
- 💉 AI-Designed Antibodies Race Toward Clinical Trials in 2026: Multiple companies expected to claim first AI-designed antibody in clinic within 1-2 years. Iambic, Generate, and Isomorphic Labs preparing to launch clinical trials. AI can now generate completely new antibody designs never before seen in nature. Industry experts predict 2026 will be first cycle where AI-native drug discovery platforms deliver clinically meaningful inflection points. Read More
- 🧫 4D Nucleome Consortium Maps 140,000+ Genome Looping Interactions: UMass Chan Medical School-led consortium mapped 3D folding of human genome over time, identifying 140,000+ looping interactions between genes and long-range regulatory elements. Data enables training deep learning models to screen DNA sequences and uncover genome folding mechanisms. Read More
- ⚖️ Pacific AI: "End of Voluntary Ethics" as Enforceable AI Law Accelerates: Throughout 2025, AI regulation shifted from principles to enforceable obligations. Overly broad definitions in EU, Colorado, and California capture virtually all software, creating "compliance splinternet" where same AI feature may be acceptable in one jurisdiction and risky in another. Read More
📚 TOP RESEARCH PAPERS
1. Evo 2: Genome modeling and design across all domains of life
Publisher: bioRxiv (Arc Institute) | Date: February 2025
Genomic foundation model trained with 40 billion parameters and 1 megabase context length on over 9.3 trillion nucleotides from 100,000+ species. Uses novel StripedHyena 2 architecture enabling training 3x faster than optimized transformers. Achieves 90% accuracy identifying disease-causing BRCA1 mutations.
Impact: For disease genetics, 90% accuracy on previously unrecognized mutations enables precise genetic counseling and personalized medicine. For synthetic biology, capacity to design entirely new genomes opens possibilities for engineering organisms with novel capabilities. The 1 megabase context window captures long-range genomic interactions previous models couldn't address.
Genomics Breakthrough
2. Pillar-0: A New Frontier for Radiology Foundation Models
Publisher: arXiv (UC Berkeley & UCSF) | Date: November 2025
Open-source AI foundation model interpreting 3D volumes directly, recognizing hundreds of conditions from single CT or MRI exam. Novel Atlas architecture is 150x faster than traditional vision transformers. Achieved .87 AUC across 350+ findings, outperforming all public competitors.
Impact: Open-source release democratizes state-of-the-art medical imaging AI for institutions without massive budgets. 150x speed improvement enables real-time clinical deployment. Ability to recognize hundreds of conditions from single exam could reduce diagnostic errors by flagging unexpected findings.
Medical Imaging AI
3. 4D Nucleome: Mapping 140,000+ Genome Looping Interactions
Publisher: 4D Nucleome Consortium (UMass Chan) | Date: January 2026
Most comprehensive map of 3D genome folding in human embryonic stem cells over time, identifying 140,000+ looping interactions between genes and regulatory elements. Enables training deep learning models to predict functional consequences of genomic variants.
Impact: Crucial for interpreting non-coding GWAS variants that exert effects by altering long-range regulatory interactions. Provides reference atlas for understanding how variants affect gene regulation. Enables identification of druggable targets and training AI models to predict variant consequences.
3D Genomics
4. Bridging AI and biology: Foundation models meet human physiology and disease
Publisher: ScienceDirect (Review) | Date: January 2026
Comprehensive review identifying critical limitation: correlation-based predictions from models trained on observational data miss causal pathways. Argues hybrid models combining deep learning with biological mechanisms will enable causal reasoning necessary for clinical decision support.
Impact: Provides roadmap for next phase of biomedical AI development. Hybrid models incorporating biological knowledge can make predictions that generalize better and provide interpretable explanations for clinicians. Addresses fundamental challenge of confusing correlation with causation.
AI/Biology Integration
💻 TOP GITHUB REPOS
1. Evo2 - Arc Institute's Genomic Foundation Model
⭐ Rapidly growing (January 2026 release)
Largest publicly available genomic foundation model with 40B parameters, trained on 9.3 trillion nucleotides across all domains of life. Predicts pathogenic variants with 90% accuracy, designs novel genomes up to 1 megabase.
Bio-Relevance: Essential for variant interpretation, synthetic biology, and genome engineering across bacteria, archaea, and eukaryotes.
⭐ 5,100+ stars (updated January 9, 2026)
Ready-to-use scientific skills for Claude covering bioinformatics, genomics, computational biology, drug discovery, proteomics, metabolomics, and chemoinformatics.
Bio-Relevance: Democratizes computational biology expertise for labs without dedicated bioinformatics teams.
⭐ 4,900+ stars
Foundational Python library for biological computation providing tools for sequence analysis, protein structure manipulation, phylogenetics, and database interfacing.
Bio-Relevance: Essential infrastructure underlying most Python-based bioinformatics pipelines.
4. claude-code - Agentic Coding Tool
Trending (January 6, 2026)
Terminal-integrated agentic coding assistant understanding full codebase context and executing multi-step development tasks via natural language.
Bio-Relevance: Enables automated bioinformatics pipeline development, code refactoring, and documentation generation.
⭐ 3,800+ stars
Google's deep learning-based variant caller using convolutional neural networks to identify genetic variants from next-generation sequencing data.
Bio-Relevance: Widely used in clinical genomics for identifying disease-causing mutations.
⭐ 3,800+ stars
Community-maintained curated list of bioinformatics libraries, software, databases, and learning resources organized by application area.
Bio-Relevance: Comprehensive entry point for researchers transitioning into computational biology.
🛠️ TOP AI PRODUCTS
1. Eli Lilly-NVIDIA AI Co-Innovation Lab
Category: Drug Discovery Infrastructure | $1B over 5 years
New AI co-innovation lab in San Francisco Bay Area combining Lilly's biology expertise with NVIDIA's AI engineering for accelerated closed-loop drug discovery. Operations begin early 2026.
Learn More
2. Evo 2 via NVIDIA BioNeMo
Category: Genomics / Foundation Model | 40B parameters
Arc Institute's Evo 2 available via NVIDIA BioNeMo, democratizing access to cutting-edge genomic AI with optimized inference and drug discovery pipeline integration.
Learn More
3. Pillar-0 Open-Source Release
Category: Medical Imaging / Open-Source AI | .87 AUC across 350+ findings
Complete codebase, trained models, and pipelines publicly released. Atlas architecture 150x faster than transformers enables real-time clinical deployment.
Learn More
4. NVIDIA-Thermo Fisher Autonomous Lab Infrastructure
Category: Laboratory Automation | Announced at JPM 2026
Partnership building autonomous lab infrastructure integrating AI from computational design through wet lab execution for closed-loop discovery systems.
Learn More
5. Iambic & Generate AI Drug Pipelines
Category: AI Drug Discovery | 3+ AI-designed drugs expected in trials 2026
Leading AI-native biotechs preparing clinical trials of AI-designed antibodies and therapeutics never before seen in nature.
Learn More
6. Isomorphic Labs Clinical Pipeline
Category: AI Drug Discovery / Structural Biology | Human trials approaching
Leveraging AlphaFold for rational drug design, preparing clinical validation of AI-designed molecules optimized for target proteins.
Learn More
⚠️ AI CRITICISM & CONCERNS
1. "The End of Voluntary Ethics": Global Shift to Enforceable AI Law
Pacific AI's 2025 policy review documents global acceleration from ethical guidelines to enforceable obligations. Overly broad definitions in EU, Colorado, and California capture virtually all software, creating "compliance splinternet" where same AI feature may be acceptable in one jurisdiction and risky in another. Healthcare AI developers face unprecedented complexity navigating divergent frameworks.
Read Analysis
2. Regulatory Overreach: Experts Predict Tide Turning Against Broad AI Regulation
Experts expect turning of tide in 2026 as EU and some U.S. states reconsider harmful effects of regulating too early and too broadly. Critics argue regulations use definitions so expansive they capture virtually all software, creating impossible compliance burdens based on hypothetical future harms rather than demonstrated present risks.
Read More
3. Federal-State Conflicts: Trump Executive Order Challenges State AI Laws
President Trump's December 2025 executive order challenges California's AI Safety Act and Colorado's algorithmic discrimination law, creating legal uncertainty about which regulations healthcare AI companies must follow. Federal-state conflict may require Supreme Court resolution of federalism questions.
Read Analysis
4. Nature Editorial: Call for International AI Safety Cooperation
Nature argues more coherence needed in global policymaking. Fragmented global response makes international collaboration increasingly difficult. Lack of coordination particularly affects healthcare, where clinical trials and medical devices operate globally, requiring duplicate validation studies that waste resources and delay patient access.
Read Editorial
💭 CLOSING REFLECTION
The second week of 2026 reveals healthcare AI's dual reality with striking clarity. We witness extraordinary technical progress: the $1 billion Eli Lilly-NVIDIA partnership, Evo 2's 40 billion parameters trained on life's genomic diversity, Pillar-0's 150x speed advantage making real-time AI diagnosis feasible, and multiple AI-designed antibodies racing toward clinical validation.
Yet as these capabilities accelerate, the governance landscape fragments further. Pacific AI documents "The End of Voluntary Ethics" as jurisdictions worldwide move from principles to enforcement, creating a "compliance splinternet." Federal-state conflicts intensify as President Trump's executive order challenges state AI laws. Nature's call for international cooperation highlights the gap between our global technical capabilities and fragmented governance frameworks.
The path forward requires simultaneous progress on two fronts. Technically, we must continue advancing AI capabilities while prioritizing interpretability, fairness, and validation. Politically, we need pragmatic governance that distinguishes real present harms from hypothetical future risks, scales oversight with actual risk levels, and coordinates across jurisdictions to reduce compliance fragmentation.
Most importantly, we need dialogue among all stakeholders. The Lilly-NVIDIA partnership shows industry confidence. Evo 2 and Pillar-0 demonstrate academic commitment to open science. Now we need policymakers, clinicians, patients, ethicists, and technologists working together to design governance frameworks that enable beneficial innovation while preventing foreseeable harms.
|