Virtual Library

Start Your Search

Helena Cirenajwis



Author of

  • +

    P2.03 - Biology (ID 162)

    • Event: WCLC 2019
    • Type: Poster Viewing in the Exhibit Hall
    • Track: Biology
    • Presentations: 1
    • Moderators:
    • Coordinates: 9/09/2019, 10:15 - 18:15, Exhibit Hall
    • +

      P2.03-02 - A Single Sample Predictor of Transcriptional Lung Adenocarcinoma Subtypes: Predicting Biology and Prognosis (ID 2201)

      10:15 - 18:15  |  Author(s): Helena Cirenajwis

      • Abstract
      • Slides

      Background

      Lung adenocarcinoma accounts for nearly 40% of all lung cancers, thereby representing the major histological subtype. Comprehensive molecular studies have proposed three molecular lung adenocarcinoma subtypes termed the terminal respiratory unit (TRU), proximal-inflammatory (PI), and the proximal-proliferative (PP) subtype based mainly on transcriptional patterns. These subtypes have been linked to molecular characteristics but also to differences in prognosis, favoring the TRU subtype compared to the PI and PP (combined PI and PP = non-TRU) subtypes. However, the method used (nearest centroid classification=NCC) to classify samples into transcriptional subtypes depends on the cohort composition, consequently struggle with e.g. reproducibility and classification of single samples. In this study, we aimed to derive a single sample predictor (SSP) of these subtypes, capable of predicting single samples irrespective of technical platforms and cohort composition.

      Method

      In this study, the gene expression based SSP called “Absolute assignment of breast cancer Intrinsic Molecular Subtype” (AIMS) were trained on a large combined dataset collection (n=1655, 17 datasets) with AC assignment obtained by the NCC method and tested in 5 publicly available gene expression datasets (n=977) with treatment data available. Survival analysis was performed to compare the two classification methods (AIMS vs. NCC) with overall survival (OS) as clinical endpoint. Survival curves were compared using Kaplan-Meier estimates and the log-rank test.

      Result

      Using AIMS, a SSP of lung adenocarcinoma transcriptional biology and prognosis was successfully trained and tested in publicly available gene expression datasets. The majority of the samples were assigned equally by the two methods. However, a minor subset (n=97) of samples were given discordant labels. The derived SSP had an accuracy of 85.5% in 977 independent validation samples for TRU vs non-TRU cases, independent of gene expression platform. Interestingly, the patient group consisting of samples classified as TRU by the NCC method and nonTRU by AIMS (TRU.nonTRU, Fig.1, orange line), showed a survival outcome more similar to the nonTRU patient group. The reverse was observed for the nonTRU.TRU patient group (Fig.1, grey line) with a survival pattern resembling that of the TRU group. Thus, on a survival basis, the discordant samples seems to be more accurately classified by the AIMS method.

      figure_1.png

      Conclusion

      We present a SSP for proposed transcriptional adenocarcinoma subtypes capable of predicting single samples irrespective of technical platform and cohort composition, thereby overcoming critical limitations in the applicability of gene signatures. The classifier provides refined categorization of patients with respect to prognosis, representing a prognostic predictor in lung adenocarcinoma.

      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.