Several knowledgebases, such as CIViC and OncoKB, have been manually curated to support clinical interpretations of a limited number of "hotspot" somatic mutations in cancer, yet discrepancies or even conflicting interpretations have been observed among these knowledgebases. Additionally, while these knowledgebases have been extremely useful, they typically cannot interpret novel mutations, which may also have functional and clinical impacts in cancer. To address these challenges, we developed an automated interpretation tool called CancerVar (Cancer Variants interpretation) to score more than 12.9 million somatic mutations and classify them into four tiers: strong clinical significance, potential clinical significance, uncertain clinical significance, and benign/likely benign, based on the AMP/ASCO/CAP 2017 guideline. Considering that the AMP/ASCO/CAP rule-based scoring system may have inherent limitations, such as lack of a clear guidance on weighing different pieces of functional evidence or unclear definition for certain clinical evidence, it may cause misinterpretation for certain variants that have functional impacts but no proven clinical significance. To address this issue, we further introduced a deep learning-based scoring system to predict oncogenicity of mutations by semi-supervised generative adversarial network (SGAN) method using both functional and clinical evidence. We trained and validated the SGAN model on 5,234 somatic mutations from an in-house database of clinical reports on cancer patients, and achieved a good performance when testing on 6,226 variants that were curated by us through literature search. We also compared the prediction with several independent datasets and showed great utility in classifying variants with previously unknown interpretations. CancerVar is also incorporated into a web server that can generate automated texts with summarized descriptive interpretations, such as diagnostic, prognostic, targeted drug responses and clinical trial information for many hotspot mutations. In summary, CancerVar can facilitate clinical interpretation and hypothesis generation for somatic mutations, and greatly reduce manual workload for retrieving relevant evidence and implementing existing guidelines.

中文翻译：

---

CancerVar：人工智能授权的平台，可对癌症中的体细胞突变进行临床解释

一些知识库（例如CIViC和OncoKB）已手动创建，以支持对癌症中有限数量的“热点”体细胞突变的临床解释，但是在这些知识库中已观察到差异甚至冲突的解释。此外，尽管这些知识库非常有用，但它们通常无法解释新颖的突变，这些突变也可能对癌症产生功能和临床影响。为了应对这些挑战，我们开发了一种自动解释工具，称为CancerVar（癌变体解释），对超过1290万个体细胞突变进行评分，并将其分为四个等级：强大的临床意义，潜在的临床意义，不确定的临床意义以及良性/可能良性，基于AMP / ASCO / CAP 2017指南。考虑到基于AMP / ASCO / CAP规则的评分系统可能具有内在的局限性，例如缺乏权衡不同功能证据的明确指南或对某些临床证据的定义不明确，因此可能会导致对具有功能的某些变体产生误解影响，但没有经过证实的临床意义。为了解决这个问题，我们进一步引入了基于深度学习的评分系统，以功能和临床证据通过半监督生成对抗网络（SGAN）方法预测突变的致癌性。我们从内部关于癌症患者临床报告的数据库中对5234个体细胞突变进行了SGAN模型的训练和验证，并且在对6226个由我们通过文献检索确定的变体进行测试时，取得了良好的性能。我们还将预测结果与几个独立的数据集进行了比较，并显示出在用以前未知的解释对变量进行分类中的巨大效用。CancerVar还集成到Web服务器中，该服务器可以生成具有摘要性描述性解释的自动化文本，例如诊断，预后，靶向药物反应以及许多热点突变的临床试验信息。总之，CancerVar可以促进体细胞突变的临床解释和假设生成，并大大减少了检索相关证据和实施现有指南的人工工作量。CancerVar还集成到Web服务器中，该服务器可以生成具有摘要性描述性解释的自动化文本，例如诊断，预后，靶向药物反应以及许多热点突变的临床试验信息。总之，CancerVar可以促进针对体细胞突变的临床解释和假设生成，并大大减少了检索相关证据和实施现有指南的人工工作量。CancerVar还集成到Web服务器中，该服务器可以生成具有摘要性描述性解释的自动化文本，例如诊断，预后，靶向药物反应以及许多热点突变的临床试验信息。总之，CancerVar可以促进体细胞突变的临床解释和假设生成，并大大减少了检索相关证据和实施现有指南的人工工作量。