Background: Following radical cystectomy (RC) for muscle-invasive bladder cancer (MIBC), recurrence is suboptimally associated with pathologic stage. A prognostic model that incorporates information from comprehensive genomic and transcriptomic data from germline and tumor tissues will improve the selection of high-risk patients (pts) for adjuvant therapy and tailored surveillance. Methods: Pts with MIBC in TCGA who underwent RC (without neoadjuvant chemotherapy) and with whole exome sequencing (WES) of tumor + normal, and tumor RNA-Seq were eligible. Germline variants were called and annotated using GATK4 and Variant Effect Predictor (VEP). Only high and moderate VEP-Impact variants contributed to burdens. Given that recurrence is associated with 5-year survival, pts were required to have data for survival at a landmark of 5 years post-RC or may have died within 5 years of RC. Patients who died <4 months of diagnosis were excluded since they may have died from post-operative complications. Dead patients were propensity-matched against alive patients by pathologic stage, adjuvant chemotherapy, gender, race, age and smoking. A linear artificial neural network was trained to predict 5-year survival using Artificial Intelligence Quality Control (AIQC)-a deep learning experience tracker. The samples were divided into 3 equally distributed subsets of alive and dead pts. The algorithm was trained on 50% of the dataset, evaluated on 40% and tested on 10%. The remaining 39 dead pts were used for additional holdout validation. The accuracy (percentage of correct prediction of death or survival) was reported. Results: A cohort of 117 pts were evaluable for 5 year survival. The cohort included 39 survivors that satisfied inclusion criteria, and 78 dead pts that were propensity matched against them (one set of 39 dead pts for training & evaluation, and a second set of 39 dead pts as an additional holdout). The binary classification neural network was trained on 38 samples to predict survival based on their most differentiated: germline alteration burdens + tumor alteration burdens + tumor gene expression. The differentially mutated genes most important to the algorithm were: PTPRN(n-m), TECTA(t-m), BPIFB1(n-m), SORL1(n-m), SLC39A5(n-m), and RBBP6(n-h) [t=tumor/ n=normal tissue origin | m=moderate/ h=high VEP impact]. The algorithm was 97% accurate on the 38 training samples, and 94% accurate across the 79 evaluation (validation, test and holdout) samples. Conclusions: Neural network analysis of germline and tumor profiling was validated to optimally differentiate between dead and surviving pts at 5 years following RC for MIBC. Further validation and application of this promising algorithm may advance precision medicine by permitting the selection of the most appropriate pts with high-risk MIBC for adjuvant therapy.