Abstract: In January 2018, Microsoft and Adaptive Biotechnologies launched the TCR-Antigen Map Project. This ambitious initiative combines immunology and machine learning to enable us to read the immune system to better diagnose and treat disease. The core of the project is the biology of T cells, which serve as the command-and-control center for the adaptive immune response. Learning is achieved through randomization and selective expansion of antigen-specific T cells, the specificity of which are determined by small stretches of amino acids on the T-cell receptor (TCR) protein. The result is that sequencing TCRs from a small blood sample yields hundreds of thousands of sequences, which collectively encode an individual’s immunological history. Over the past decade, Adaptive has built out the technology to generate unprecedented quantities of patient-level labeled training data (sets of TCRs with labels associated with “disease”) and specific TCR-antigen pairs (individual TCRs labeled with their antigen specificity). The goal of the project is to develop the experimental and computational tools and techniques to develop a blood test for the early and accurate detection of many diseases. This technology may also be applied to predict patient outcomes from blood samples, further enhancing its use as a diagnostic tool, and to rapidly identify the specific T-cell-antigen interactions that lead to good outcomes, which could potentially inform therapeutics and vaccines. In principle, these approaches could have broad applicability across cancers, autoimmune disorders, and infectious diseases. By early 2020, our primary clinical achievement was a candidate Lyme disease diagnostic, which has entered clinical development.

Then COVID-19 happened. In late March, we extended our collaboration to focus on COVID-19, taking the tools we’d been developing for two years and applying them to the emerging pandemic. We have collected thousands of blood samples to map the T-cell response to the SARS-CoV-2 virus to understand how T cells do their job. We are curating this data in what we call the ImmuneCODE database and making it publicly available for researchers around the world to help develop solutions. The aim of the project is to develop a deeper understanding of how immune responses correlate with disease outcomes—in effect, letting the immune system show us how to control infection. Adaptive is also applying these data to develop a diagnostic test for patients based on the T-cell response to the virus as well as a research tool to monitor vaccines in development or to guide development of second generation vaccines.

In this talk, I’ll give an overview of the project and how we’re approaching reading the immune system as a machine learning problem, then describe some of our recent results applying these tools to COVID-19.

Bio: Jonathan Carlson, PhD, is a senior director in Microsoft’s Health Futures group. After majoring in biology and computer science as an undergraduate at Dartmouth, he completed his PhD in Computer Science at the University of Washington, where he was coadvised by Larry Ruzzo (UW) and David Heckerman (Microsoft Research). In 2008, he joined eScience group at Microsoft Research, where he studied how human viruses evolve around immune-mediated selection pressures, with the a specific focus on HIV. In 2016, he joined Project Premonition, where he led the development of cloud-scale metagenomics approaches, and in 2018, founded the TCR-Antigen Map Project with Adaptive. More details can be found on his website.