Author of

• 34780

  +

  P37 - Pathology - Biomarker Testing (ID 107)

  • Event: WCLC 2020
  • Type: Posters
  • Track: Pathology, Molecular Pathology and Diagnostic Biomarkers
  • Presentations: 1
  • Moderators:
  • Coordinates: 1/28/2021, 00:00 - 00:00, ePoster Hall

  • 34804

    +

    P37.35 - Identification of DNA Methylation Markers to Distinguish Early-Stage Lung Adenocarcinomas from Benign Pulmonary Nodules (ID 2231)

    00:00 - 00:00  |  Presenting Author(s): Feilong Zhao
    • Abstract
    • Slides

    Introduction
    

    Low-dose CT screening could identify pulmonary nodules yet is unable to distinguish the benign one. Previous studies have discovered lots of DNA methylation biomarkers to solve this problem, however, most of these studies screened only a panel of genes and generated limited specificity. We aimed to screen biomarkers more extensively to identify biomarkers that could improve the predicting specificity.

    Methods
    

    26 lung adenocarcinomas of stage I/II and 22 benign pulmonary nodules up to 30mm in diameter were subjected to Gentron-health 180-gene mutation sequencing and capture-based methylation sequencing which covered 5.5M CpGs positions. A random-forest classifier was trained using this dataset. A validating set composed of 16 lung adenocarcinomas of stage I/II and 14 benign pulmonary nodules and a testing set composed of 16 lung adenocarcinomas of stage I/II and 15 benign pulmonary nodules were collected afterward.

    Results
    

    A majority of patients were male and never smokers. All of the malignant samples were invasive adenocarcinoma, while the benign ones were composed of inflammation nodules, inflammatory granuloma, fungal infection, and fibrosis. Density of malignant nodules included solid (n=10), mGGN (n=11) and pGGN (n=5), while that of benign ones included solid (n=19) and mGGN (n=4). The most frequent mutations detected were EGFR and TP53. Differentially methylated positions (DMPs) were filtered using the following criteria: 1). With adjusted p-value less than 0.05; 2). Locate in the promoter region of cancer-driving genes. 3). Correlated with the expression of the corresponding gene in the TCGA-LUAD dataset. Afterward, features composed of 285 DMPs and the mutation status of EGFR or TP53 were further filtered by recursive feature elimination using a random-forest classifier, and 6 features produced the highest nested-cross-validation score in the training set were selected. A random-forest classifier based on them generated a performance of 95.83% in the training set. Genes where the 6 features locate in were PREX1, GGTLC1, MEST, CLEC14A, and LTC4S. As the enrolled patients were all nodule harboring, our foremost goal is to identify the benign ones, i.e. to improve the specificity of the classifier. So we evaluated the prediction probability threshold in an independent validating set, which demonstrated the default threshold (probability > 0.5) created the best specificity. The random-forest classifier was then tested in an independent testing set, resulting in an accuracy of 87.10%, a sensitivity of 81.25%, and a specificity of 93.33%.

    Conclusion
    

    We identified 6 CpGs and created a random-forest classifier using them, which could distinguish early-stage lung adenocarcinomas from benign pulmonary nodules with a sensitivity of 81.25% under a specificity of 93.33%. As all of them locate in promoter regions of cancer-driving genes, the 6 CpGs are well worth further validating in blood.

    Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.