CellNeighbor：患者肿瘤与细胞系模型的转录图谱，用于指导临床前模型选择

Preclinical scientists typically select cell line models for compound activity assays based on molecular profiles that support a drug program's hypothesis. However, extensive in vitro culturing of these cell lines can influence genetic changes in these models raising concerns about whether these models still represent the patient tumors from which they were derived. To address this translational gap, we developed CellNeighbor, a computational framework that guides model selection by contextualizing cell lines within the landscape of real-world patient transcriptomic data. Building on the established Celligner framework (1), we explicitly identified and removed tumor microenvironment-related gene expression variability before integrating cell line and patient profiles with Harmony (2). Then, we integrated transcriptomic profiles from DepMap (3,4) with patient tumor data from The Cancer Genome Atlas (TCGA) and deidentified patient tumor data from Tempus, creating a unified transcriptomic map of cell lines and tumors. Within this map, we employ a symmetric nearest neighbor approach to identify "patient neighborhoods" centered around each cell line to characterize cell lines most transcriptionally similar to patient tumors. In addition, we have developed novel metrics, such as a neighborhood-tissue homogeneity score, to quantify the confidence of these associations for use in objective and automated decision making. We can now rank cell lines by their transcriptional similarity to neighborhoods of real patient tumors to increase the likelihood that model lines remain representative of the intended patient population. The integrated clinical and molecular data layers produce a cell line-to-patient map that enables cell line selection tailored to specific patient populations. In conclusion, CellNeighbor offers a robust method to identify cell line models that closely resemble patient tumors, ultimately aiming to increase the translatability of preclinical discoveries into clinical applications. 1.Warren, A., Chen, Y., Jones, A. et al. Global computational alignment of tumor and cell line transcriptional profiles. Nat Commun 12, 22 (2021). https://doi.org/10.1038/s41467-020-20294-x 2.Korsunsky, I., Millard, N., Fan, J. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods 16, 1289-1296 (2019). https://doi.org/10.1038/s41592-019-0619-0 3.DepMap, Broad (2025). DepMap Public 25Q3. Dataset. depmap.org 4.Arafeh, R., Shibue, T., Dempster, J.M. et al. The present and future of the Cancer Dependency Map. Nat Rev Cancer 25, 59-73 (2025). https://doi.org/10.1038/s41568-024-00763-x