3.2.1. Breast Cancer

Metabric is a breast cancer study that involved more than 2000 patients with primary breast tumors [26]. The data set provides copy number aberration, gene expression, mutation and long-term clinical follow-up information. We are interested in the clinical variable of tumor grade, which measures the abnormality of the tumor cells compared to normal cells under a microscope. It takes a value of 1, 2, or 3. Higher Grade indicates more abnormality and higher risk of rapid tumor proliferation. Since grade 1 contained the fewest samples, we pooled it together with Grade 2 as one class and treated Grade 3 as the other class.

We then applied PKB to samples in subtype Lum B, where the sample sizes for the two classes were most balanced (259 Grade 3 patients; 211 Grade 1,2 patients). For input gene expression data, we used the normalized mRNA expression (microarray) data for 24,368 genes provided in the data set. The model using GO Biological Process pathways and radial basis function (rbf) kernel yielded the best performance (error rate 27.4%). To obtain the pathways most relevant to tumor grade, we calculated the average pathway weights from the cross validation and sorted them from highest to lowest. Top fifteen pathways with the highest weights are presented in the first columns of Table 4.

Among all pathways, the cell aggregation and sequestering of metal ion pathways are the top two pathways in terms of the estimated pathway weights. Previous research has shown that cell aggregation contributes to the inhibition of cell death and anoikis-resistance, thereby promoting tumor cell proliferation. Genes in the cell aggregation pathway include TGFB2, MAPK14, FGF4 and FGF6, which play important roles in the regulation of cell differentiation and fate [27]. Moreover, the majority of genes in the sequestering of metal ion pathway encode calcium-binding proteins, which regulates calcium level and different cell signaling pathways relevant to tumorigenesis and progression [28]. Among these genes, S100A8 and S100A9 have been identified as novel diagnostic markers of human cancer [29]. The results suggest that PKB has identified pathways that are likely relevant to breast cancer grade.
