Uploaded May 2026 | Updated September 2026, 2 weeks ago
Eric and Wendy Schmidt Center Symposium: Biomedical Science and AI
April 28 - 29, 2026
Day 1, Short talk: Synthetic Survival Control: Extending Synthetic Controls for “When-If” Decision in Healthcare
Jessy Xinyi Han, Devavrat Shah Lab, MIT
For more information, visit broad.io/ewsc-symposium-2026
Copyright Broad Institute, 2026. All rights reserved.
Eric and Wendy Schmidt Center Symposium: Biomedical Science and AI
April 28 - 29, 2026
Day 1, Short talk: Synthetic Survival Control: Extending Synthetic Controls for “When-If” Decision in Healthcare
Jessy Xinyi Han, Devavrat Shah Lab, MIT
For more information, visit broad.io/ewsc-symposium-2026
Copyright Broad Institute, 2026. All rights reserved.



![EWSC: Engineering sharper cancer immunotherapies using robotics and machine learning
Schmidt Center - MIT EECS Joint Colloquium Series
Presented by the Eric and Wendy Schmidt Center
March 9, 2026
Broad Institute of MIT and Harvard
Engineering sharper cancer immunotherapies using robotics and machine learning
Colloquium with Grégoire Altan-Bonnet
Deputy Chief, Laboratory of Integrative Cancer Immunology,
National Cancer Institute
Abstract:
Cancer immunotherapies elicit highly variable outcomes in patients and genetically identical mouse models, suggesting a strong intrinsic stochastic component. Using thousands of well-controlled ex vivo immunoassays, we show that leukocyte activation and tumor cytotoxicity display macroscopic variability that follows a shifted Poisson distribution. This variability arises from stochastic activation of a rare subpopulation of T cells (“Spark T cells”) coupled to a paracrine IFN-γ–driven positive feedback. By integrating these quantitative insights into a custom machine-learning pipeline with single-cell resolution, we phenotypically and functionally identify Spark T cells in murine naïve T cells and in human T cell blasts used for adoptive cell therapy, and demonstrate their role in shaping heterogeneous immunotherapy responses [http://dx.doi.org/10.2139/ssrn.4996071]. Building on this framework, we investigate how receptor-level signal integration modulates T cell potency and specificity. While chimeric antigen receptor (CAR) T cells show strong cytotoxicity but limited specificity in solid tumors, endogenous T cell receptors (TCRs) provide exquisite antigen discrimination with reduced efficacy. Deploying our high-throughput platform and mathematical modeling [PMID: 35587980], we reveal inhibitory and cooperative crosstalk between co-expressed TCRs and CARs: strong TCR–antigen interactions enhance CAR activation, whereas weak interactions antagonize it. Leveraging this crosstalk, we engineer dual TCR/CAR T cells targeting neoantigens and HER2 that exhibit enhanced antitumor activity with minimal off-tumor toxicity in a humanized solid tumor mouse model [PMID: 40220754] Together, these results show how stochastic T cell activation and receptor-level signal integration jointly govern variability, potency, and precision in cancer immunotherapy.
Questions? Email Amanda Ogden at aogden@broadinstitute.org.
Copyright Broad Institute, 2026. All rights reserved. EWSC: Engineering sharper cancer immunotherapies using robotics and machine learning](https://i.ytimg.com/vi/WzUd2_pQcyI/mqdefault.jpg)






