Biobanking as Enabling Infrastructure for AI-Based Predictive Oncology

High-quality, rigorously characterized biospecimens and associated multimodal clinical data are an indispensable foundation for reliable oncology prediction models.  Biobanks that stress the need for standardized procurement procedures, annotation, quality control, and openness in multimodal phenotyping information as well histopathological analysis, molecular analysis, and clinical endpoints, offer a crucial enabling infrastructure for the latest advances, especially in the use of AI algorithms for extracting information from tumors, inferring a signature, and predicting a response to therapy that has not yet been approached in its accuracy. 
Objectives:
This study underscores the pivotal, foundational role of high-quality biobanked cohorts, in particular the Molecular Twin breast cancer cohort, in empowering AI-driven prediction models for immunotherapy response. It stresses the paramount importance of a deep understanding of cohort characteristics, proactive anticipation of future AI applications, and strategic leveraging of AI to derive actionable insights directly from routine histopathology, thereby accelerating progress in predictive oncology for diverse cancers and therapies.
Methods:
The Oncobiobank’s Molecular Twin-driven breast cancer collection comprises well-characterized, high-quality formalin-fixed, paraffin-embedded (FFPE) samples with H&E-stained slides, collected between January 2013 and July 2024. They are also associated with comprehensive and standardized clinical data (such as treatments and RECIST response). Derived from Cedars-Sinai’s “Molecular Twin Research Umbrella Protocol” (study IRB STUDY00001879), these resources enabled the external validation of models based on deep learning methods such as Path2Omics, which can predict informative transcriptomic-based signatures (such as TIME.ACT for tumor immune activation) based exclusively on routine histopathology slides without the need for expensive molecular studies. The validation was performed on pre-therapeutic samples from patients treated with immune checkpoint blockade (ICB), which is sometimes combined with chemotherapy or targeted therapies.
Results:
This advanced invasive breast carcinoma cohort, with its carefully annotated data, was crucial for enabling the first-ever external validation of AI-induced transcriptomic scores. The predictions from the slides demonstrated strong discriminative power for response to ICB, which surprisingly approached the gold-standard RNA-based metrics and was far superior to traditional immune signatures in cross-validation across multiple cohorts. The above findings clearly illustrate the important role of high-quality, well-annotated biospecimens, as well as the enabling role of biobanks in enabling the revolutionary identification of “hot” tumors by AI without the need for comprehensive molecular analysis.
 Conclusion: The above findings clearly demonstrate that biobanks serve not only as passive infrastructure but also as a foundation for future-prone high-end technological oncology research by actively utilizing new technologies to realize the value of prospective cohorts for new tasks. The use of AI in quality biobanks would enable more predictive analyses, enabling rapid translation of basic to applied research for precision oncology. This further reinforces the significant potential of high-quality biobanks as a foundation for translating scientific breakthroughs into life-saving cancer control for all forms of cancer.