Archives

  • 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • 2021-03
  • br In brief we first

    2019-09-16


    In brief, we first applied data from half of the METABRIC cohort advantages over Integrative Cluster, which was originally proposed by
    (training set, n = 952) to a machine learning algorithm known as a ran- the provider of the METABRIC data set [19]. The stratification based on
    dom forest classifier and thereby selected 23 genes. We optimized the mPS is thus more significant than that based on Integrative Cluster
    weight for each gene with a neural network algorithm. We thus built (Fig. 3, a and c), and mPS is less expensive to apply than Integrative
    a molecular prognostic score (mPS) that is calculated by summation of Cluster, for which whole-genome sequence analysis is required.
    Fig. 2. Identification of all prognosis-related genes in the TCGA breast cancer cohort and validation in 36 independent cohorts. (a) Distribution of MYC AngiotensinI level (RSEM) among patients in the TCGA breast cancer cohort (left), and Kaplan-Meier curves of OS for these patients based on a MYC expression level higher or lower than the median (right). The HR, its 95% CI, and the log-rank P value are shown. (b) Kaplan-Meier curves of OS for the TCGA cohort based on MKI67 expression level. (c and d) Kaplan-Meier curves of OS for the TCGA cohort based on TMEM65 (c) and ENOSF1 (d) expression levels, respectively. The complete list of OS-related genes in this TCGA discovery cohort is provided in Supplementary Table S2. (e) Logarithm of the integrated HR for all 184 prognosis-related genes in the validation data sets. The complete list of these genes identified by meta-analysis is provided in Supplementary Table S3.
    Please cite this article as: H. Shimizu and K.I. Nakayama, A 23 gene–based molecular prognostic score precisely predicts overall survival of breast cancer pati..., EBioMedicine, https://doi.org/10.1016/j.ebiom.2019.07.046
    Fig. 3. mPS precisely stratifies prognosis of breast cancer patients. (a) Kaplan-Meier curves of OS according to mPS for the METABRIC test cohort. (b) Kaplan-Meier curves of OS according to PAM50 classification for the METABRIC test cohort. We omitted one patient whose PAM50 classification was not available. (c) Kaplan-Meier curves of OS for the METABRIC test cohort according to Integrative Cluster, which was proposed by the provider of the METABRIC data set. (d) Kaplan-Meier curves of OS according to mPS for the public data set GSE86166. (e) Kaplan-Meier curves of OS for the GSE86166 data set according to the 12-chemokine gene expression score proposed by the provider of the data set. (f) Kaplan-Meier curves of OS according to mPS for the public data set GSE96058.
    Fig. 4. mPS AngiotensinI is applicable to most breast cancer subsets. (a) Kaplan-Meier curves according to mPS for OS of patients with claudin-low tumours in the METABRIC test cohort. (b) Kaplan-Meier curves according to mPS for OS of premenopausal patients (b50 years of age) in the METABRIC test cohort. (c) Kaplan-Meier curves according to mPS for OS of patients in the METABRIC test cohort with ILC. (d and e) Kaplan-Meier curves according to mPS for OS of patients in the METABRIC test (d) and TCGA (e) cohorts at clinical TNM stage II. (f) Kaplan-Meier curves according to mPS for OS of patients in the Moderate II cluster of the NPI in the METABRIC test cohort.
    Table 2
    Univariate and multivariate analyses of OS in the TCGA cohort. The HR relative to the indi-cated reference (ref) value, its 95% CI, and P value (those of b0.05 are indicated in bold) for the Cox hazard model are shown.
    Univariate
    Multivariate
    ratio
    ratio
    Age
    Gender
    Stage
    mPS
    Many existing prognostic indicators are able to predict prognosis on the basis on only one specific platform or pipeline. We overcame this limitation by changing continuous gene expression values (which may vary depending on method) to discrete values (high or low relative to the median), rendering mPS independent of platform. For demonstra-tion purposes, we analyzed another breast cancer data set, GSE86166, in which transcriptome profiling was performed by microarray analysis [22]. We found that mPS stratifies OS into different bins in the same way as in the METABRIC test cohort (Fig. 3d), showing that mPS is applicable to both RNA-sequencing–based (METABRIC) and microarray-based (GSE86166) data sets. In addition, the mPS system is superior to the 12-chemokine gene expression score [22] proposed by the provider of the GSE86166 data set (Fig. 3, d and e).