From Emergent AI Insights to Chrono- Immunotherapy: Learning When to Treat Cancer

Abstract

Immunotherapy efficacy is increasingly understood to depend not only on what treatment is given, but also when it is administered. At the same time, advanced AI offers a powerful opportunity to uncover clinically relevant immunological patterns that may escape conventional hypothesis-driven analysis. In this lecture, I will elucidate how our AI framework, initially trained to capture neoantigen-specific HLA-I repertoire and co-associated spatially-defined myeloid-T cell interactions, showed an unexpected emergent learning capability. It independently identified circadian immune biology as a clinically relevant determinant of immunotherapy response in kidney cancer patients. This discovery prompted a chrono-immunotherapy program showing strong promise for morning-timed immune checkpoint blockade in kidney cancer patients that uncovered unexpected chrono-immunobiology associated with immunotherapy-deterministic myeloid-T cell interactions. The concept has now progressed into the ongoing TIC-TAC study, a treatment-matched, historical cohort-controlled observational single-arm study testing whether time-optimized immunotherapy can translate AI-derived biological insight into better patient care.