Machine Learning for Healthcare and Life Sciences, IBM Research - Haifa, Mount Carmel Campus, Israel.
Computational Biology Center, IBM T. J. Watson Research, Yorktown Heights, NY 10598, USA.
Sci Rep. 2016 Dec 23;6:38988. doi: 10.1038/srep38988.
Compiling a comprehensive list of cancer driver genes is imperative for oncology diagnostics and drug development. While driver genes are typically discovered by analysis of tumor genomes, infrequently mutated driver genes often evade detection due to limited sample sizes. Here, we address sample size limitations by integrating tumor genomics data with a wide spectrum of gene-specific properties to search for rare drivers, functionally classify them, and detect features characteristic of driver genes. We show that our approach, CAnceR geNe similarity-based Annotator and Finder (CARNAF), enables detection of potentially novel drivers that eluded over a dozen pan-cancer/multi-tumor type studies. In particular, feature analysis reveals a highly concentrated pool of known and putative tumor suppressors among the <1% of genes that encode very large, chromatin-regulating proteins. Thus, our study highlights the need for deeper characterization of very large, epigenetic regulators in the context of cancer causality.
编译一份全面的癌症驱动基因清单对于肿瘤学诊断和药物开发至关重要。虽然驱动基因通常通过分析肿瘤基因组来发现，但由于样本量有限，罕见的突变驱动基因往往难以检测到。在这里，我们通过将肿瘤基因组学数据与广泛的基因特异性特性相结合来解决样本量限制的问题，以寻找罕见的驱动基因，对其进行功能分类，并检测驱动基因的特征。我们表明，我们的方法 CAnceR geNe similarity-based Annotator and Finder (CARNAF) 能够检测到十几个泛癌/多肿瘤类型研究中遗漏的潜在新驱动基因。特别是，特征分析揭示了编码非常大的染色质调节蛋白的基因中 <1%的基因中存在高度集中的已知和假定的肿瘤抑制因子。因此，我们的研究强调了在癌症因果关系的背景下，需要更深入地研究非常大的、表观遗传调节因子。
