AI-Simulated in Silico Humans: A New Paradigm for Modeling Autoimmunity

Building AI-powered digital avatars that simulate each patient's immune system to predict therapy response and accelerate precision immunotherapy development.
Ongoing Other Academia–Industry Partnerships Clinical Trials Collaboration & Innovation Data-Driven & Quantitative Experimental Platforms & Models Immune Profiling Machine Learning & AI Precision Medicine Single Cell Technologies Systems Biology Translational & Clinical

Funding Year: 2024

Project Overview

Autoimmune diseases affect over 50 million Americans and require therapies that target immune dysfunction without broad suppression — but traditional lab models often fail to capture the complexity of human biology. This project, developed in close partnership with Google Research and Google DeepMind, uses large language models and the foundational Cell2Sentence approach to build in silico "human avatars" that simulate each patient's immune system at the single-cell level. By converting patient immune data into a language AI can interpret, the platform can predict how different therapies may impact inflammation and tissue damage before patients enter a trial — enabling faster, safer development of precision immunotherapies tailored to individual patients.

Impact & Innovation

Digital twins for autoimmune drug discovery.

 

In partnership with Google Research and DeepMind, this project builds AI avatars that simulate individual immune systems at single-cell resolution — enabling virtual trials, therapy prediction, and disease mechanism discovery before patients are ever enrolled.

  • Enables prediction of individual therapy response and inflammation outcomes directly from patient immune data, accelerating drug discovery and reducing trial risk
  • Generates IP through a patent-pending technology (US 2025/0139386 A1) built on the peer-reviewed Cell2Sentence model, with a collaborative open-source approach extending its reach
  • Advances the Consortium’s Integrated Data and Discovery Platforms pillar by pioneering AI-driven digital avatars as a replicable infrastructure for precision autoimmune research across institutions
Research Approach

A framework designed for discovery

This project combines large-scale AI language modeling with single-cell patient immune data to construct and validate in silico human avatars capable of simulating immune system behavior and predicting therapeutic response in autoimmune disease. The work builds on the peer-reviewed Cell2Sentence framework in close partnership with Google Research and Google DeepMind.

Application of the C2S-Scale large language model and Cell2Sentence approach to convert single-cell patient immune data into AI-interpretable language; construction of in silico human avatars simulating individual immune systems at single-cell resolution; and virtual therapy testing to predict inflammation and tissue damage outcomes prior to clinical trials.

Single-cell patient immune data from autoimmune disease cohorts, the C2S-Scale large language model (bioRxiv 2025), the foundational Cell2Sentence framework (ICML 2024), and collaborative datasets developed in partnership with Google Research and Google DeepMind.

Development and validation of AI-powered digital avatars that predict individual patient responses to precision immunotherapies, with translational goals including enabling virtual clinical trials, accelerating targeted treatment development, and providing clinicians and industry partners with actionable, individualized predictions. The platform is patent-pending (US 2025/0139386 A1) and designed to serve as a replicable model for AI-driven autoimmune research.

Investigators & Institutions

Powering the science

Principal Investigator

David van Dijk, PhD, MSc, BSc, Colton Consortium Member

Associate Professor, Department of Internal Medicine (Cardiology), Yale School of Medicine, Yale University

Colton Centers & Institutions

Colton Center for Autoimmunity at Yale University

Research Outputs

From insight to impact

  • Peer-reviewed publication: Cell2Sentence, ICML 2024. 
  • Preprint: C2S-Scale large language model, bioRxiv 2025. Research partnership with Google Research and Google DeepMind.
  • Patent-pending technology: US 2025/0139386 A1.