IMMUNOLOGY2025™ Conference Recordings-Navigating the high-dimensional space of cancer immunotherapies

Video Transcription

Video Summary

The session opened with welcomes from the co-chairs, Catherine Blish and John Tsang, who announced poster-session “AI ambassadors” to help attendees, especially first-timers. The first speaker, Grégoire-Otten Barnet, described how machine learning and systems immunology can improve cancer immunotherapy.  <br /><br />He first showed that high-dimensional CyTOF profiling of T-cell products used in adoptive cell therapy can predict patient outcomes. Simple immune signatures, such as CD39-low/CD69-low cells, were linked to better tumor regression and validated in mouse models.  <br /><br />He then introduced a second framework for studying why T-cell killing is so variable in vitro. By running highly controlled tumor–T cell assays, his group found “shifted Poissonian” noise, suggesting that a rare subset of “SPARC” T cells initiates strong immune responses. These cells, marked by CD5-low/CD11A-high phenotypes, appear poised for rapid interferon-gamma production and can be traced back through fate mapping and chromatin accessibility analyses. Their signature was also predictive in melanoma patients with low MHC tumors receiving checkpoint blockade.  <br /><br />Finally, he described a machine-learning model of T-cell activation built from large, robotically generated cytokine datasets. This model revealed a low-dimensional “latent space” for antigen response and was used to design a new CAR-T strategy that combines CAR and TCR signaling to improve tumor killing while reducing toxicity in healthy tissue.

Keywords

machine learning

systems immunology

cancer immunotherapy

CyTOF profiling

T-cell activation

SPARC T cells

CAR-T therapy