Predicting response to cisplatin-based neoadjuvant chemotherapy for muscle-invasive bladder cancer: transcriptomic features outrank genomic biomarkers

Muscle-invasive bladder cancer (MIBC) is associated with poor predictability of response to cisplatin-based neoadjuvant chemotherapy (NAC). Consequently, the benefit of NAC remains unclear for many patients due to the lack of reliable biomarkers. In order to identify biomarkers and build an integrated and accurate model to predict NAC response, we conducted a comprehensive transcriptomic and genomic profiling on 100 MIBC patient tumors. Our analysis identified 602 differentially expressed genes between responders and non responders. From these, we derived the Top10up and Top10dn gene signatures. We also identified a set of genes related to keratinization (KRT), extracellular matrix (ECM) and cell cycle regulation (CELL CYCLE REG) that were strongly associated with response. Additionally, we found a Wnt-related signature (WNT) significantly associated with non response. Genomic analysis revealed that mutations in DNAH8, DNAH6 or DNAH10 (DNAHalt) and deletions in KDM6A (KDM6Adel) correlated significantly with chemotherapy resistance. Using our comprehensive molecular analysis as a backbone, we developed different machine learning (ML) models, based on the Xgboost (XGB) approach. The transcriptomics-only model (XGB-R) matched the performance of the combined model (XGB-RW; AUC=0.85). The relative ease of sample collection for transcriptomic data, along with the external validation, make it a promising candidate for clinical translation. To ensure robustness and broader applicability, external validation in larger, more diverse cohorts is necessary. To facilitate this, our most predictive ML models have been made publicly available via GitHub. Statement of significanceWe present XGB-R, a robust validated machine-learning model using transcriptomic features to predict neoadjuvant chemotherapy response in muscle-invasive bladder cancer with high accuracy (AUC=0.85), being a candidate for clinical translation.