Your browser does not support JavaScript!
2017-02-17(五),主講人:謝叔蓉 教授 (中央研究院 統計學研究所)


統 計 學 研 究 所

專 題 演 講


講 題: Detection of Somatic Mutations in Exome Sequencing of Tumor-only Samples
演講者: 謝叔蓉教授 (中央研究院 統計學研究所)
時 間: 106年02月17日(星期五)10:40 - 12:00noon  (10:20 - 10:40am 茶會於統計所821室舉行)
地 點: 綜合三館837室
摘 要:


Recently next generation sequencing data has been used in practice for cancer research and clinic diagnosis. Quite a few algorithms have been developed to detect somatic mutations for matched tumor-normal patient samples. However, to reduce diagnosis costs and due to lack of normal samples, detecting small mutations from tumor-only sample is required but relatively unexplored. To this end, we have developed an algorithm (GATKcan) based on GATK for detecting mutations of cancer. In addition to four statistics of GATK hard filtering (resulted from a pilot study on endometrial cancer), GATKcan incorporates (1) two statistics to characterize and remove false mutations, (2) optimization and machine learning methods to train the thresholds of the statistics using small fraction of reported mutations of endometrial cancer in TCGA in each of ten repeats.


The averaged performance of GATKcan of the ten repeats was better than GATK in detecting mutations of the remaining 231 endometrial tumors. As external validations, we further applied the three algorithms to TCGA breast cancer (BC), ovarian cancer (OC) and melanoma tumors, GATKcan was shown to outperform GATK in detecting mutations of 27,167 to 50,799 called variants in these cancers, treating TPR and FPR equally important. Importantly, GATKcan reduced high fractions of false positives detected by GATK in the four cancers, while validation is costly (US$5-10 per variant in Taiwan). GATKcan also outperformed GATK and VarScan 2 in mutation detection of somatic variants classified by VarScan 2 in BC and (was equivalent to VarScan 2) in melanoma. GATKcan performed slightly worse than VarScan 2 in OC, which may be due to ~30% of indels in called variants of OC but only ~0.8% of indels in EC. Nevertheless, GATKcan does not require normal samples, thus it reduces sequencing costs to half, and enables detection of mutations while alternate alleles exist in normal samples, which remains a bottleneck in the area.


These results indicate that GATKcan can be trained by exome-seq of another cancer similarly, and is expected to detect mutations of future patients with the same type of cancer well and is likely to be applicable to other cancers. GATKcan is expected to receive a large number of applications such as sequence-based clinical diagnoses.

This is a joint work with Yu-Chin Hsu, Yu-Ting Hsiao, and Jan-Gowth Chang


敬請公佈                                 歡迎參加

  • 友善列印
  • 分享
Voice Play