A Pathway-Based Classification of Breast Cancer Integrating Data on Differentially Expressed Genes, Copy Number Variations and MicroRNA Target Genes

Abstract

Breast cancer is a clinically heterogeneous disease characterized by distinct molecular aberrations. Understanding the heterogeneity and identifying subgroups of breast cancer are essential to improving diagnoses and predicting therapeutic responses. In this paper, we propose a classification scheme for breast cancer which integrates data on differentially expressed genes (DEGs), copy number variations (CNVs) and microRNAs (miRNAs)-regulated mRNAs. Pathway information based on the estimation of molecular pathway activity is also applied as a postprocessor to optimize the classifier. A total of 250 malignant breast tumors were analyzed by k-means clustering based on the patterns of the expression profiles of 215 intrinsic genes, and the classification performances were compared with existing breast cancer classifiers including the BluePrint and the 625-gene classifier. We show that a classification scheme which incorporates pathway information with various genetic variations achieves better performance than classifiers based on the expression levels of individual genes, and propose that the identified signature serves as a basic tool for identifying rational therapeutic opportunities for breast cancer patients.

Pathway-based classification of breast cancer subtypes

“…Three recent methods, i.e., (63)(64)(65), instead, focus on a similar problem. In this regard, the performance of our method is definitely comparable to that shown in (63) in which, however, the authors make use also of the additional information on Copy number variations and microRNAs-regulated mRNAs to reduce the number of features (i.e., total accuracy slightly lower than 90%, sensitivity values ranging between 0.5.2 and 0.9.8 according to the different subtypes, and specificity values between 0.8.5 and 0.9.6, see Figure 6 in (63)).…”

“…Three recent methods, i.e., (63)(64)(65), instead, focus on a similar problem. In this regard, the performance of our method is definitely comparable to that shown in (63) in which, however, the authors make use also of the additional information on Copy number variations and microRNAs-regulated mRNAs to reduce the number of features (i.e., total accuracy slightly lower than 90%, sensitivity values ranging between 0.5.2 and 0.9.8 according to the different subtypes, and specificity values between 0.8.5 and 0.9.6, see Figure 6 in (63)). With respect to (64) we obtain a considerably better overall accuracy, yet they focus on the classification of metastatic/non metastatic breast tumors, which is a notably different problem (i.e., accuracy values significantly lower than 70% with respect to the two considered datasets and all the tested methods, see Figure 4 in (64)).…”

Integrating genetics and epigenetics in breast cancer: biological insights, experimental, computational methods and therapeutic potential

“…[ 397 – 399 ]). Eo et al [ 391 ] proposed a pathway-based classification of BC which integrates data on DE genes, CNA and miRNA. Pathway information was incorporated in a condition-specific manner.…”

RNA-Seq-Based Breast Cancer Subtypes Classification Using Machine Learning Approaches

“…We believe that the different groups will certainly hold distinct molecular interaction mechanisms, so we constructed the gene coexpression networks with weighted DEGs based on Pearson correlation coefficients (PCC) for control and experiment groups, respectively. There is a lot of work [ 21 – 23 ] that can use Pearson correlation coefficients to build a coexpression network based on high-throughput FPKM data from TCGA database. In addition, in order to meet the requirements of the normal distribution, we performed some data preprocessing operations, such as log2 transformation.…”