TMEs-Atlas: An Open Multi-Modal Database of Human Tumor Tissue States and Perturbation Responses to Catalyze AI-Driven Immunotherapy Design

Curing cancer requires strategies that move beyond single-gene or single-pathway targeting to approaches that harness human data to address the complex ecosystems sustaining tumors. While past advances have identified oncogenic drivers, therapeutic success in patients has been limited by models that fail to capture the diversity and stability of tumor microenvironments (TMEs). Tumors evade immunity not as isolated cells but as recurrent, archetypal ecosystems that resist therapy. Understanding and reprogramming these archetypes is therefore an urgent challenge. We propose that this challenge represents a unique opportunity for integrating experimental and AI-driven approaches to uncover the cellular and molecular programs that define stable tumor ecosystems and test how they can be shifted from cancer-promoting to cancer-eliminating states. Our central hypothesis is that gene-expression programs and ecosystem-level interactions can be systematically mapped and perturbed to reveal the most effective therapeutic interventions. Achieving this requires two complementary directions: (i) high-dimensional data collection to capture cellular networks and their responses to diverse perturbations, and (ii) computational models to define archetypal states and predict transitions. Together, these directions establish a generalizable framework for understanding and manipulating tumor ecosystems, exemplifying how AI can accelerate discovery and therapy design in complex biological systems.