目前高通量定序資料已廣泛的運用在肺腺癌的研究中，包括表觀基因組學，基因體學和轉錄組學等等；全面分子特徵有助於確定生物標誌物以及用於早期診斷與治療。因應精準醫療的臨床需求，目前次世代定序隨著高通量定序品質提高以及成本的降低，使得其應用在臨床的可行性大幅提高。本篇論文使用基因體圖譜計畫（TCGA）中的肺腺癌基因組數據。我們採用生物統計分析以及深度學習的方法，分析TCGA肺腺癌多種基因組數據。我們整合TCGA肺腺癌372位病人包括RNA、miRNA、甲基化基因組數據，建立預測因子模型來預測肺腺癌高風險和低風險患者疾病進展存活。在統計分析中找出十七基因印記可顯著預測患者疾病進展存活，並在另外三組肺腺癌病人中驗證所找出的基因印記能準確預測存活。在深度學習分析方面，我們透過自編碼器深度學習模型建立預測模型並在驗證組上有良好的預測效果。;High-throughput genomic assays have been widely used to investigate lung adenocarcinoma by employing technologies of epigenomics, genomics and transcriptomics. Comprehensive characterization of molecular mechanisms has contributed to identify biomarkers for early diagnosis and treatment. Nowadays, high-throughput sequencing technology providing a higher coverage and lower cost is prevalent in clinical application. This study used biostatistical and deep learning-based computing methodologies to conduct data mining and modeling on adenocarcinoma lung cancer datasets extracted from TCGA data. We integrated RNA, miRNA, and DNA methylation genomic data from 372 lung adenocarcinoma patients. Based on biostatistical methods, we developed a 17-gene signature to distinguish the risk groups with disease-free survival. In addition, we validated a 17-gene signature in three independent lung adenocarcinoma cohorts. Besides statistical validation, we also used Deep Learning-based autoencoder modeling to validate these datasets.
