用于癌症药物筛选数据可重复分析的计算框架

Background: Screening therapeutic responses in patient-derived or model-based cancer cells provides a direct experimental strategy to evaluate treatment efficacy. These assays generate complex datasets that require scalable, standardized analysis pipelines to ensure reproducibility and cross-study comparability. However, current data management and analysis workflows remain fragmented, relying on manual curation and ad hoc scripts that hinder reproducibility. This study introduces an open-source computational framework that standardizes storage, analysis, and visualization of high-throughput screening (HTS) data, providing easy access to common analysis methods while maintaining reproducibility in clinical and academic settings. Methods: We developed a Python framework that provides a coherent workflow for storing, processing, and analyzing high-throughput drug screening data. It enables efficient and uniform storage with consistent data structure across experiments and datasets. The framework integrates automated drug name standardization, standardized data preprocessing, and common dose-response modeling methods including IC50, EC50, and DSS calculations. In addition, the framework supports cohort-level summarization and treatment prioritization, allowing users to compare drug responses across patients. Together, these components create a reproducible and adaptable pipeline that transfers raw experimental measurements to interpretable biological and translational insights. Results: To demonstrate its broad utility, we applied the framework to three use cases. With the imported GDSC2 dataset, our recomputed IC50 and AUC values were highly consistent with published data, while providing integrated visualization and more contextual insights. We re-analyzed a breast cancer PDX-derived organoid dataset and were able to evaluate batch effects and compare responses across multiple models. We also analyzed an ex-vivo drug screening data from an ongoing pediatric brain tumor precision oncology initiative. The outputs were used to aid in prioritizing treatment candidates for individual patients at molecular tumor board meetings, demonstrating applicability in translational contexts. Across all cases, the framework ensured consistent preprocessing, minimized manual data manipulation, and provided comprehensive cross patient and cross drug comparison for identifying personalized drug candidates. Conclusion: Our framework provides a comprehensive, interoperable solution for managing cancer drug screening data. By minimizing manual data manipulation and enabling reproducible analysis, it bridges the gap between experimental data generation and actionable insight. Beyond personalized ex-vivo assays, this platform empowers systematic analysis across public and clinical datasets, facilitating translational research and personalized treatment selection.