Stratifying renal carcinoma patients using data-driven immune & epithelial landscapes

Renal cell carcinoma (RCC) ranks in the top ten most frequent cancers worldwide and is responsible for approximately 134,000 deaths per year. Clear-cell RCC (ccRCC), which accounts for 70–80% of cases, is characterized by genomic instability and dysregulation of normal cell-to-cell communication in a spatially aware manner. Although ccRCC tumors are known to be immunogenic, patient response has remained highly variable and no reliable molecular markers for patient treatment selection have entered the clinic thus far. Proposed biomarkers, such as tumor mutational burden, have consistently failed to predict ICB response. Similarly, existing transcriptomics-driven tumor classification systems identify 4 to 6 distinct subtypes but have yet to provide a reliable basis for therapy response prediction. This research seeks to address this critical gap in predicting therapy response by designing a data-driven cancer subtype classification system. To ensure the robust nature of this classification system, methods will be rigorously benchmarked against existing subtyping strategies, testing their performance across patient cohorts, and reviewed by subject matter experts. This benchmarking will involve comparing precision, recall, and overall predictive power in identifying therapy-sensitive subtypes. Methods to combine data across modalities will also be developed in this work, ensuring insights are transferable between analyses. To achieve these goals, we will work with both public and in-house datasets that include single-cell and bulk-resolution omics data from ccRCC patients. Our approach will focus on several key subtasks:   • Omics integration and information transfer:   • Bulk sample deconvolution • Functional module summarization • Building of novel classification models Through this work, we aim to deliver a cutting-edge, semi-supervised framework for renal carcinoma subtyping, building on existing clustering methods for predicting ICB sensitivity, with the potential to inform future clinical trial designs.