一种量子驱动的图谱规模资源，用于研究癌症TME中的代谢图景与异质性

Metabolic networks are profoundly rewired in cancer, supporting uncontrolled proliferation, immune evasion, and therapy resistance. Although large atlas-level single-cell RNA-seq (scRNA-seq) datasets of cancer tumor microenvironments (TMEs) and normal human tissues are now available, there remains no systematic, genome-scale characterization of metabolic landscapes, shifts, heterogeneity, and cross-talk across cancer and normal cells using these data. Elementary flux modes (EFMs), representing the minimal sets of reactions that sustain steady-state flux, provide a principled basis for describing metabolic capabilities, yet identifying biologically feasible EFMs in genome-scale networks is a major bottleneck due to the combinatorial explosion of possible modes. We propose a quantum-enabled framework that leverages atlas-level scRNA-seq data to infer plausible EFM distributions and derive sample-specific metabolic fluxes, with the long-term goal of building a scalable resource of cancer metabolic landscapes. Both EFM discovery and flux prediction are formulated as Quadratic Unconstrained Binary Optimization (QUBO) problems solved on quantum annealing hardware. By exploiting the intrinsic parallel sampling capabilities of quantum devices, our approach efficiently explores high-dimensional solution spaces under stoichiometric, thermodynamic, and cancer-specific biological constraints. To handle genome-scale models relevant to oncology, we integrate tensor decomposition-based dimensionality reduction to yield tractable QUBO formulations while preserving key pathway structure. On simulated networks with up to 25 reactions, quantum sampling robustly enriches EFMs that satisfy stoichiometric balance, support minimality, and irreducibility, while structurally invalid modes are rarely sampled. When imposing distinct sample-specific constraints mimicking different tumor contexts, the high-frequency EFM sets shift systematically, demonstrating that the framework captures condition-specific flux distributions without explicit enumeration. We then apply this quantum-driven strategy to single-cell atlases of pancreatic, prostate, and breast cancer, and integrate these results with TCGA bulk RNA-seq in constraint-based models to generate a shared, extensible resource of cancer metabolic landscapes. We show that this resource can be used to map recurrent pathway vulnerabilities, prioritize metabolic intervention targets, and quantify condition-specific metabolic strategies within and across tumor types, treatment states, and patient cohorts. By bridging quantum optimization and systems biology, our work provides a practical and interpretable foundation for personalized metabolic modeling and hypothesis generation in cancer.