Abstract 2878: BCAM (breast cancer attractor metagenes): A new tool for assessing breast cancer prognosis

• Published in: Cancer research (Chicago, Ill.) Ill.), 2014-10, Vol.74 (19_Supplement), p.2878-2878
• Main Authors: Cheng, Wei-Yi, Yang, Tai-Hsien Ou, Maurer, Matthew, Anastassiou, Dimitris
• Format: Article
• Language: English
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2014-2878

The winning model of the Sage Bionetworks/DREAM Breast Cancer Prognosis Challenge (Sci Transl Med, 17 April 2013: Vol. 5, Issue 181, p. 181ra50) made use of several novel molecular features, called attractor metagenes, as well as another metagene involving the expression levels of two genes, FGD3 and SUSD3, which are genomically adjacent to each other. Here we present the results of our follow-up work developing a breast cancer prognostic test, called BCAM (Breast Cancer Attractor Metagenes). BCAM was derived from the Challenge winning models by excluding unusable features and optimizing performance in predicting breast cancer specific survival. BCAM incorporates underlying tumor biology by including five molecular features: the FGD3-SUSD3 metagene and four attractor metagenes (CIN, MES, LYM, and HER2, which are associated with mitotic chromosomal instability, mesenchymal transition, lymphocyte infiltration, and expression of the HER2 amplicon, respectively); as well as incorporating the extent of disease: tumor size and the number of positive lymph nodes. Based on analysis of three breast cancer data sets with appropriate whole transcriptome and clinical outcomes data (allowing for time to recurrence as a phenotype), our results suggest that the combination of features used in BCAM outperforms the combination of features used in existing breast cancer biomarker products: Oncotype DX, Mammaprint and PAM50. The molecular features in BCAM were also shown to have improved performance against the Oncotype DX and PAM50 features in the subset of ER positive tumors treated with hormonal therapy, and against MammaPrint in the subset of lymph node negative tumors with size less than 50 mm. In addition, performance was significantly improved when the “ER group” of the Oncotype DX panel was replaced by the FGD3-SUSD3 metagene. All evaluations of prognostic performance were shown to be statistically significant by multiple rounds of random splitting. We also present a web-based version of BCAM (http://128.59.65.24:8080/brcabiomarker), in which uploaded gene expression data from a patient's tumor are analyzed and integrated with tumor size and number of positive nodes, and a report is generated containing a percentile prognostic score against the 2,000 patient METABRIC data set (Nature, 21 June 2012: Vol. 486, Issue 7403, p. 346-52), the corresponding ten-year breast cancer specific survival rate, and additional scores representing individual molecular features. We cur