Abstract 7641: Prognostic impact of functional germline mutations in muscle-invasive bladder cancer identified by neural network analysis

Abstract

Abstract Background: An accurate prognostic algorithm based on germline whole exome sequencing (WES) would complement tumor omics as a non-invasive source of biomarkers for guiding cancer patient therapy. In a previous study, we used WES and RNA-seq data from muscle-invasive bladder cancer (MIBC) and normal host cells to train a survival model, wherein we discovered that mutated germline genes accounted for 90% of feature importance. We have since defined a larger cohort to identify key germline biomarkers of overall survival (OS) that may be scalable for application in the clinic. Methods: 169 patients from The Cancer Genome Atlas (TCGA) with MIBC stages T2-T4 that underwent radical cystectomy (RC) w/o neoadjuvant treatment were eligible. OS at 4 years was the primary clinical endpoint as it is associated with death due to metastatic tumor recurrence. Patients were dichotomized to control survivors who were alive at 4+ years, and cases who died within 4 years but no less than 4 months after RC. 49 qualifying survivors were propensity matched against 49 dead patients by stage, adjuvant treatment, gender, race, and age. The remaining 71 dead patients served as a holdout. Only high (e.g. frameshift) and moderate (e.g. indel) impact germline variants were included. Mutations were weighted for allele frequency (penalize common) and genotype (reward homozygous) before grouping at the gene level. A panel of the 21 most differential genes in stratified patients were selected for neural network training using AIQC. The model forced protective/pathogenic genes to interact and produced individual survival probabilities. Gene importance was determined via permutation. Results: OS was accurately predicted for 98.2% of patients: 1 FN, 2 FP, 37 TN, 119 TP. The top 3 permuted genes (and related pathways) were: SORL1 (Aβ), KIF27 (hedgehog), CUL7 (p53). The most differential gene was CUL7: 31% of survivors mutated vs 2% of dead. Conclusion: Neural network analysis identified the promise of germline genetics as an accurate predictor of survival and tumor recurrence in MIBC patients. Upon validation in an external cohort, these biomarkers could be used to inform adjuvant therapy and complement molecular residual disease alongside circulating tumor (ct)-DNA. Citation Format: Guru P. Sonpavde, Amin H. Nassar, Arvind Ravi, Layne Sadler. Prognostic impact of functional germline mutations in muscle-invasive bladder cancer identified by neural network analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 7641.