| 引用本文: | 张巧生,许俊杰,韦亚龙,孙振宇,张恒,仲兆满.基于拓展通路的胶质母细胞瘤[]预后预测[J].生物信息学,2025,23(4):313-322. |
| ZHANG Qiaosheng,XU Junjie,WEI Yalong,SUN Zhenyu,ZHANG Heng,ZHONG Zhaoman.Prognostic prediction of glioblastoma based on extended pathways[J].Chinese Journal of Bioinformatics,2025,23(4):313-322. |
|
| 摘要: |
| 胶质母细胞瘤(Glioblastoma,GBM)是颅内侵袭性最强、预后效果最差的原发恶性肿瘤,因为表达数据的高维度特性导致预后预测方面的结果精度很低。构建能够解决高维、低样本量数据的鲁棒计算问题的预后预测模型对GBM的医学研究具有积极意义,并且这一领域目前仍未有较好的解决方案。本研究提出了一种拓展通路关联的深度神经网络(EPDNN),使用基于图论的拓展基因通路的方式,将紧密的调节基因加入到通路中,使模型能够学习更多特征,然后,通过整合条件生成对抗网络进行数据增强,最后根据预测结果评估模型性能。五折交叉验证后,EPDNN相比传统预后预测分类器取得了最高的曲线下面积和F1分数,也优于当前最新的预后预测模型PASNet,为GBM的个体化术后治疗提供了指导治疗工具。同时模型能够直观地表示基因和通路的分层关系及其非线性关系,在深度学习的可解释性研究上作出了探索。 |
| 关键词: 胶质母细胞瘤 深度神经网络 平衡数据集 拓展通路 预后预测 |
| DOI：10.12113/202409012 |
| 分类号:TP183 |
| 文献标识码:A |
| 基金项目:国家自然科学基金项目(No. 72174079);连云港市科技项目(No. CG2223,No. CG2323)；连云港市博士后基金资助项目(No. LYG20210010). |
|
| Prognostic prediction of glioblastoma based on extended pathways |
|
ZHANG Qiaosheng,XU Junjie,WEI Yalong,SUN Zhenyu,ZHANG Heng,ZHONG Zhaoman
|
|
(School of Computer Engineering, Jiangsu Ocean University, Lianyungang 222000, Jiangsu China)
|
| Abstract: |
| Glioblastoma (GBM) is the most aggressive intracranial primary malignant tumor with the worst prognosis, because the high-dimensional nature of the expression data leads to low accuracy of the results in terms of prognosis prediction. Constructing prognostic prediction models that can solve the robust computational problem for high-dimensional, low-sample-volume data is of positive significance for medical research on GBM, and there is still no better solution in this field. In this study, we proposed an extended pathway-associated deep neural network (EPDNN), which used a graph-theory-based extension of gene pathways by adding tightly regulated genes to the pathways to enable the model to learn more features, and then, data augmentation was carried out by integrating conditional generative adversarial networks, and finally, the performance of the model was evaluated based on the prediction results. After five-fold cross-validation, EPDNN achieved the highest area under the curve (AUC) and F1 scores compared to traditional prognostic prediction classifiers, and the tight genes identified in the extended pathway stage of the model were identified as important genes for GBM in previous biological and medical studies.The EPDNN model outperformed the current state-of-the-art prognostic prediction models, and provided a guided therapy for individualized postoperative treatment of GBM Tools. Meanwhile, the model was able to intuitively represent the hierarchical relationship between genes and pathways and their nonlinear relationship, which maked an exploration on the interpretability study of deep learning. |
| Key words: Glioblastoma Deep neural network Balanced dataset Extended pathway Prognostic prediction |
