AI drug discovery, cancer prediction, and smarter hospitals.
A newsletter on the latest in AI for healthcare.
Welcome back,
This week, Isomorphic Labs raised $2.1 billion to expand its AI-driven drug discovery platform, while Compass, a new foundation model from the Zitnik Lab at Harvard Medical School, showed how tumour gene activity data can help predict which cancer patients may respond to immunotherapy.
On the technical side, the top AI model for health is the StanfordMIMI/Merlin model, which continues to push medical imaging AI forward by helping interpret scans and clinical information together.
Here is what you need to know today,
Compass uses tumour gene activity data to help predict which cancer patients may respond to immunotherapy.
Isomorphic Labs raises $2.1 billion to expand its AI-powered drug discovery platform.
Merlin is an AI model from the StanfordMIMI team that can analyse medical images and clinical information together.
Startup rounds, deals, and moves.
IKS Health, a healthcare technology company focused on clinical, administrative, and revenue cycle support, acquires ARAI Solutions to strengthen its healthcare AI platform
Shyld AI raises $13.4M to expand its hospital AI tools, including systems that use UV-C light to disinfect rooms and reduce infection risk without adding extra work for clinical staff
Quick reads across health AI.
FDA approves early warning system for sepsis - AI is being used to warn clinicians earlier when a patient may be developing sepsis.
UChicago Medicine to Implement Smart Hospital Platform - AI is being used to improve hospital workflows, including operating room coordination and patient care processes.
Source: medRxiv (preprint) · 01 Jan 2026
This paper introduces Compass, a pan-cancer foundation model designed to predict patient response to Immune Checkpoint Inhibitors (ICI). By moving beyond single-gene biomarkers to a transcriptome-wide approach, the model addresses the high failure rate of immunotherapy in unstratified cohorts.
This work helps clinicians and researchers identify which patients are most likely to benefit from specific treatments across diverse cancer types.
Research Question
Can a foundation model trained on diverse tumour transcriptomes accurately predict immunotherapy outcomes across multiple cancer types and treatment regimens?
Approach
Study design: Retrospective development and validation of a foundation model using transcriptome data, which shows which genes are switched on or off in tumour cells.
Architecture: Compass uses a concept-bottleneck transformer, meaning it turns tumour gene activity into interpretable immune concepts before predicting treatment response.
Data: Pre-trained on over 20,000 pan-cancer transcriptomes and fine-tuned on dedicated immunotherapy cohorts.
Population: Patients across multiple cancer types (e.g., melanoma, lung, bladder) receiving various ICI treatments.
Endpoints: Objective response rate (ORR), progression-free survival (PFS), and overall survival (OS).
Results
Compass significantly outperformed traditional biomarkers, in predicting treatment response.
The model demonstrated strong generalisability, maintaining predictive power across independent datasets and diverse malignancies.
High-risk scores were strongly correlated with shorter progression-free survival across multiple treatment lines.
Potential impact: Compass provides a framework for precision oncology, potentially reducing unnecessary toxicity and costs by refining patient selection for immunotherapy.
Source: PR Newswire · 15 May 2026
Isomorphic Labs has closed a massive $2.1 billion Series B funding round to accelerate its AI-first approach to drug discovery. The company, which emerged from Google DeepMind, intends to use the capital to expand its therapeutic pipeline and refine its proprietary "next-generation" biological models. The funding marks one of the largest private investments in the AI biotech sector to date.
Image from Isomorphic Labs
Platform Expansion: Scaling the AI drug design engine to handle increasingly complex multi-target proteins.
Pipeline Growth: Moving internal programmes closer to clinical candidate selection across oncology and immunology.
Infrastructure: Significant investment in compute resources and high-throughput experimental labs to create a closed-loop "dry-to-wet" lab environment.
Why it matters: This investment shows that investors are making bigger bets on AI-aided drug design as a major part of the future of pharmaceutical R&D, with significant promise for developing new medicines for diseases that remain difficult to treat.
Source: StanfordMIMI · May 2026
Merlin is a multimodal foundation model specifically engineered for medical imaging tasks. Developed by the Stanford Multi-Modal Medical Imaging (MIMI) group, it integrates visual encoders with large language models to enable advanced reasoning over scans, such as identifying anomalies or generating preliminary findings from complex radiological data.
What stands out:
Multimodal Reasoning: Capable of processing 2D and 3D imaging data alongside clinical text.
Zero-shot Capabilities: Demonstrates strong performance on novel imaging benchmarks without extensive task-specific fine-tuning.
Medical Alignment: Pre-trained on curated medical datasets to reduce the hallucination risks common in general-purpose models.
Startup rounds, deals, and moves in healthcare AI.
IKS Health provides technology and services for clinical, administrative, and revenue cycle work in healthcare. It acquired ARAI Solutions to strengthen its healthcare AI platform. The goal is to use AI agents to reduce routine workload for care teams.
Shyld AI builds autonomous AI systems for hospitals. Its tools use computer vision and UV-C light to disinfect rooms, reduce contamination, and support infection control. The company raised $13.4 million to expand deployment across U.S. hospitals.
MD and PhD Researcher at Cambridge University applying physics-informed ML/AI to neurophysiological research.
May 25, 2026 · 6 min
May 23, 2026 · 6 min
May 20, 2026 · 7 min