Background: Neuroblastoma is one of the most devastating forms of childhood cancer. Despite large amounts of attempts in precise survival prediction in neuroblastoma, the prediction efficacy remains to be improved.
Methods: Here, we applied a deep-learning (DL) model with the attention mechanism to predict survivals in neuroblastoma. We utilized 2 groups of features separated from 172 genes, to train 2 deep neural networks and combined them by the attention mechanism.
Results: This classifier could accurately predict survivals, with areas under the curve of receiver operating characteristic (ROC) curves and time-dependent ROC reaching 0.968 and 0.974 in the training set respectively. The accuracy of the model was further confirmed in a validation cohort. Importantly, the two feature groups were mapped to two groups of patients, which were prognostic in Kaplan-Meier curves. Biological analyses showed that they exhibited diverse molecular backgrounds which could be linked to the prognosis of the patients.
Conclusions: In this study, we applied artificial intelligence methods to improve the accuracy of neuroblastoma survival prediction based on gene expression and provide explanations for better understanding of the molecular mechanisms underlying neuroblastoma.
Keywords: deep-learning (DL); individual therapy; neuroblastoma; survival; transcriptome.
Copyright © 2021 Feng, Xiang, Yi, Meng, Chu, Huang, Zhao, Chen, Xiong and Feng.