IBCN 2023: Gene Expression-Based Classification of Urothelial Carcinoma For Research And Clinical Applications
An initial 5-class (Uro, GU, Basal, Mes-like, and ScNE) Lund Taxonomy RF classifier was trained on 307 samples with microarray gene-expression data (preprocessed with both RMA and SCAN using fixed BrainArray version 25 annotations). A new model was trained after identification and removal of outlier samples, which was then applied to two additional uniformly preprocessed microarray datasets and 265 RNA-sequenced samples (summarized to TPM using both Kallisto and Salmon). Prediction results were in good agreement with both previous subtype classifications of these cohorts and with immunohistochemistry-based subtype assignments. A final predictor was built using all four datasets (each with two preprocessing versions) as training data. A separate classifier for substratification of Uro into UroA, UroB, and UroC was designed in a similar manner.
Application of the classifier to the TCGA, IMvigor210, UC-Genome, and Robertson et al. T1 datasets gave confident subtype predictions which recapitulated the expression patterns of the training cohort. A significant enrichment of FGFR3 and RB1 mutations were seen in UroA+UroB and GU+ScNE respectively.
The rule-based single sample classification approach performed well across multiple external datasets with minimal preprocessing requirements. Evaluation across additional datasets (RNA-seq and microarrays) is ongoing, with final validation planned in a cohort of prospectively RNA-sequenced tumors with paired IHC.
Presented by: Elena AramendÃa, Division of Oncology, Department of Clinical Sciences, Lund University, Lund, Sweden
Written by: Stephen B. Williams, MD, MBA, MS @SWilliams_MD on Twitter during the International Bladder Cancer Network (IBCN) Annual Meeting, September 29-30, 2023, Montreal, Canada