Zhao Jianmei, Li Xuecang, Yao Qianlan, Li Meng, Zhang Jian, Ai Bo, Liu Wei, Wang Qiuyu, Feng Chenchen, Liu Yuejuan, Bai Xuefeng, Song Chao, Li Shang, Li Enmin, Xu Liyan, Li Chunquan
School of Medical Informatics, Daqing Campus, Harbin Medical University, Daqing, 163319, China.
The Key Laboratory of Molecular Biology for High Cancer Incidence Coastal Chaoshan Area, Shantou University Medical College, Shantou, 515041, China.
Oncotarget. 2016 Sep 20;7(38):61054-61068. doi: 10.18632/oncotarget.11064.
While gene fusions have been increasingly detected by next-generation sequencing (NGS) technologies based methods in human cancers, these methods have limitations in identifying driver fusions. In addition, the existing methods to identify driver gene fusions ignored the specificity among different cancers or only considered their local rather than global topology features in networks. Here, we proposed a novel network-based method, called RWCFusion, to identify phenotype-specific cancer driver gene fusions. To evaluate its performance, we used leave-one-out cross-validation in 35 cancers and achieved a high AUC value 0.925 for overall cancers and an average 0.929 for signal cancer. Furthermore, we classified 35 cancers into two classes: haematological and solid, of which the haematological got a highly AUC which is up to 0.968. Finally, we applied RWCFusion to breast cancer and found that top 13 gene fusions, such as BCAS3-BCAS4, NOTCH-NUP214, MED13-BCAS3 and CARM-SMARCA4, have been previously proved to be drivers for breast cancer. Additionally, 8 among the top 10 of the remaining candidate gene fusions, such as SULF2-ZNF217, MED1-ACSF2, and ACACA-STAC2, were inferred to be potential driver gene fusions of breast cancer by us.
虽然下一代测序（NGS）技术已越来越多地检测到人类癌症中的基因融合，但这些方法在识别驱动融合方面存在局限性。此外，现有的识别驱动基因融合的方法忽略了不同癌症之间的特异性，或者仅考虑网络中的局部而非全局拓扑特征。在此，我们提出了一种基于网络的新方法，称为RWCFusion，以识别特定表型的癌症驱动基因融合。为了评估其性能，我们在35种癌症中使用留一法交叉验证，总体癌症的AUC值达到0.925，信号癌症的平均AUC值为0.929。此外，我们将35种癌症分为两类：血液学癌症和实体癌症，其中血液学癌症的AUC值高达0.968。最后，我们将RWCFusion应用于乳腺癌，发现前13种基因融合，如BCAS3-BCAS4、NOTCH-NUP214、MED13-BCAS3和CARM-SMARCA4，先前已被证明是乳腺癌的驱动因素。此外，其余候选基因融合的前10名中有8种，如SULF2-ZNF217、MED1-ACSF2和ACACA-STAC2，被我们推断为乳腺癌的潜在驱动基因融合。
