Machine learning developed immune-related exosome signature for prognosis and immunotherapy benefit in bladder cancer.

BACKGROUND

Bladder cancer is one of the most common malignancies with high invasion and poor clinical outcome. Exosomes exert a vital role in tumor development, drug resistance, and immunotherapy response.

METHODS

Based on the datasets from TCGA, GSE13507, GSE31684, GSE32984 and GSE48276, immune-related exosome signature (IES) was developed with an integrative analysis procedure containing 10 machine learning methods. To investigate the performance of IES in predicting the immunotherapy benefit, three immunotherapy datasets (GSE91061, GSE78220 and IMvigor210) and several predicting scores were used.

RESULTS

The RSF + Enet (alpha = 0.2) algorithm-based signature was considered as the optimal IES as it had a highest average C-index of 0.75. The IES presented a powerful performance in predicting the survival outcome of bladder cancer patients and their AUC of 1-, 3- and 5-year ROC curve was 0.711, 0.751 and 0.806 in TCGA dataset. A lower level of immune-activated cells and immune-related function, higher tumor immune dysfunction and exclusion score, higher immune escape score, higher intratumor heterogeneity score and lower PD1&CTLA4 immunophenoscore, and lower tumor mutational burden score were obtained in bladder cancer with high IES score, suggesting less immunotherapy benefits. Moreover, bladder cancer cases with high IES score had a higher cancer related hallmark score.

CONCLUSION

The current study developed an optimal IES in bladder cancer, which acted as an indicator for predicting clinical outcome and immunotherapy benefits for bladder cancer patients.