Phenomic Immune-Health Profiling through Self-Supervised Representation Learning

Abstract

High-content immune-cell imaging offers unprecedented opportunities to study immune aging and therapeutic response, yet its scale and complexity demand advanced computational approaches. This talk presents a self-supervised phenomics framework that learns biologically meaningful cellular representations from microscopy images without reliance on manual annotations. Using contrastive learning and attention-based aggregation, the approach identifies age-associated immune cell populations, captures donor-specific immune signatures, and characterizes patient-dependent drug perturbation responses. The Chronotype platform is introduced as an integrative application, combining phenotypic embeddings with functional and molecular data to estimate immunological age and immune resilience. Case studies from healthy donors and clinical cohorts illustrate the translational potential of phenomic profiling for immune health assessment, drug discovery, and personalized immune rejuvenation. The work underscores the role of self-supervised learning as a foundational technology for scalable and unbiased immune phenotyping.