Medea is an AI agent that nominates therapeutic targets across cell type contexts, predicts synthetic lethality in cancer cell lines, and forecasts immunotherapy response from multimodal patient profiles. To test whether Medea can identify the genetic vulnerabilities that DNA-damaging treatments exploit, we use it to scan 238,046 gene combinations for synthetic lethality. [Project website]

COMPASS is a pan-cancer foundation model that predicts immunotherapy response from tumor microenvironments and highlights the biology driving that response. [Nature Medicine paper] [Harvard Medicine News]

Treatment reasoning underpins every therapeutic decision in medicine. ATHENA an AI agent for treatment reasoning across all FDA approved drugs since 1939, trained by reinforcement learning over a universe of 212 biomedical tools. [Project website]

MedLog is an open protocol for event-level logging of medical AI, validated across four real-world pilots in the US, Switzerland, and Vietnam to enable auditing, monitoring, and governance of AI systems. [Paper] [Project website]

Biological reasoning models combine large language models with models of biological data, including DNA, RNA, and proteins. New preprint on training and evaluating 100+ biological reasoning models.

Introducing AutoScientists - self-organizing agent teams for long-running scientific experimentation. [Paper] [Code]

Three papers accepted to ICML 2026: (1) Evaluating relational reasoning in LLMs with REL, (2) Adaptive time series reasoning via segment selection, and (3) SPATIA: Multimodal model for prediction and generation of spatial cell phenotypes.

Our work on agentic AI for science featured in Nature Methods. This piece features: (1) ToolUniverse — an open platform enabling AI agents to use scientific tools and databases at scale, (2) ClawInstitute — shared research boards for long-running collaborative discovery where agents co-develop ideas over time, and (3) Medea — an omics AI agent for large-scale biological reasoning and analysis.

OptimusKG brings biomedical knowledge into a modern multimodal knowledge graph. It supports graph AI, knowledge-grounded retrieval with large language models, and discovery workflows that generate and evaluate biomedical hypotheses.

New preprint on exploring relational reasoning capabilities in LLMs with REL. [Project website] [Kempner Institute News]

ARK is an AI agent for autonomous knowledge graph exploration with adaptive breadth-depth retrieval. [ARK Agent CLI]

Excited to launch Open AI Scientists, our initiative to empower scientific discovery with AI scientists. [https://www.openscientist.ai]

Review on Generalist biological artificial intelligence in modeling the language of life in Nature Biotechnology.

Claw Institute is a research exchange for AI scientists. It gives agents a shared space to publish ideas, challenge claims, use scientific tools, and build on one another’s work. These early interactions point to a new mode of discovery in which societies of AI scientists participate in discovery loops alongside human researchers.

Five papers accepted to ICLR 2026: (1) Multi-scale protein representation learning, (2) Protein structure tokenization, (3) Knowledge-based approaches for medical reasoning, (4) Sequence models for generating biological and health trajectories, and (5) Multimodal language models for time series.

Our research has been recognized with the 2026 Overton Prize.

New preprint on the safe use of medical foundation models for real-world decision making. [Project website]

New preprint on multimodal reasoning model for time series.

Introducing STRAND, a generative model for predicting how DNA sequence-resolved perturbations shift single-cell gene expression states.

New paper on scaling medical AI across clinical contexts in Nature Medicine.