机制一致的细胞系选择：连接真实世界肿瘤驱动因素与体外模型以进行靶点验证

Cancer cell lines remain workhorses for mechanistic studies and preclinical drug prioritization, but their transcriptional programs can shift as they move from in vivo to in vitro conditions and drift under prolonged culture. Selection for growth advantage and stress tolerance can alter lineage and resistance programs, yet these models are still used to infer mechanisms of action and project efficacy to patients, often without a rigorous link to patient-level disease biology. We leveraged multimodal real-world data comprising longitudinal clinical records, tumor DNA and RNA sequencing, to infer in vivo driver mechanisms that impact outcomes under standard of care. We then developed an integrative framework that connects these patient-level mechanisms to in vitro systems where they can be perturbed and pharmacologically modulated. For each disease context, we construct predictive network models of cancer progression and identify cell type-localized “mechanism programs” whose activity associates with survival and progression. In parallel, we build matched network models for cancer cell lines using multiomic and perturbation data, deriving corresponding mechanism programs in vitro. The key step is establishing mechanistic and phenotypic concordance between patients and models. For each patient-inferred driver mechanism, we search the in vitro network space for subnetworks with similar structure and activity, defining mechanism-concordant lines and tissue cultures. In phenotypically concordant systems, perturbing the mechanism program shifts molecular state and cell viability in a manner consistent with the direction of the patient-level outcome association; in discordant systems, the same program either fails to respond or drives phenotypes inconsistent with clinical benefit, revealing culture-induced artifacts. We illustrate this framework in ovarian cancer. Multimodal real-world data reveal distinct driver mechanisms across the disease course, including hormone signaling, lineage plasticity, and platinum resistance networks. A restricted subset of ovarian cell lines and ex vivo cultures are mechanistically and phenotypically concordant for these programs; in these models, genetic or drug perturbation of implicated subnetworks reduces viability, whereas non-concordant models show limited effects. Future work aims to prospectively test modulators nominated by the integrated system in patient-derived ex vivo cultures and assess whether agents targeting concordant mechanisms shift molecular programs and viability as predicted, while agents predicted to be inactive do not. This patient-to-model mechanism bridge enables systematic selection of appropriate in vitro systems, prioritization of targets with both outcome-level and perturbation support, and calibration of preclinical effect sizes against clinically meaningful benefit.