Detection of gene pathways with predictive power for breast cancer prognosis.

Ma S, Kosorok MR. · 2010

Read Detection of gene pathways with predictive power for breast cancer prognosis. by Ma S, Kosorok MR. (2010) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc2837025.txt.

Introduction

Ma and Kosorok proposed a new way in 2010 to find gene pathways, sets of genes working together, that predict breast cancer outcome. Single-gene hunts miss the coordination between genes, and older pathway methods judged fit rather than true forecasting power. Their method scores a pathway by how well it predicts relapse-free survival, using the joint effect of all its genes at once.

The method can also pool several studies with unlike designs, which older tools could not do. Run on four breast cancer prognosis datasets, it flagged ninety-seven pathways with real predictive power. Some of those pathways had been missed entirely by rival methods. The authors offer the tool as a sharper route to prognosis clues hiding in gene groups.

Catalog record

CREATORMa S, Kosorok MR.
YEAR2010
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC2837025

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Detection of gene pathways with predictive power for breast cancer prognosis By Shuangge Ma, Michael R Kosorok Published in: BMC Bioinformatics DOI: 10.1186/1471-2105-11-1 ABSTRACT Abstract Background Prognosis is of critical interest in breast cancer research. Biomedical studies suggest that genomic measurements may have independent predictive power for prognosis. Gene profiling studies have been conducted to search for predictive genomic measurements. Genes have the inherent pathway structure, where pathways are composed of multiple genes with coordinated functions. The goal of this study is to identify gene pathways with predictive power for breast cancer prognosis. Since our goal is fundamentally different from that of existing studies, a new pathway analysis method is proposed. Results The new method advances beyond existing alternatives along the following aspects. First, it can assess the predictive power of gene pathways, whereas existing methods tend to focus on model fitting accuracy only. Second, it can account for the joint effects of multiple genes in a pathway, whereas existing methods tend to focus on the marginal effects of genes. Third, it can accommodate multiple heterogeneous datasets, whereas existing methods analyze a single dataset only. We analyze four breast cancer prognosis studies and identify 97 pathways with significant predictive power for prognosis. Important pathways missed by alternative methods are identified. Conclusions The proposed method provides a useful alternative to existing pathway analysis methods. Identified pathways can provide further insights into breast cancer prognosis. BACKGROUND Amongst women in the US, breast cancer is the most commonly diagnosed malignancy after skin cancer, and is the second leading cause of cancer deaths after lung cancer. According to the American Cancer Society, in 2009, an estimated 192,370 new cases of breast cancer were diagnosed, and 40,610 died from breast cancer. Women in the US have a 1 in 8 lifetime risk of developing invasive breast cancer and a 1 in 33 overall chance of dying from it. Biomedical studies suggest that genomic measurements may have independent predictive power for breast cancer prognosis [1,2]. Multiple gene profiling studies have been conducted, searching for genomic measurements with predictive power for breast cancer prognosis. "Breast cancer has probably been the carcinoma most intensively studied by gene expression profiling" [1]. In this article, when referring to "prognosis", we limit ourselves to relapse-free survival. The overall and other types of survival have different patterns and different genomic bases, and need to be investigated separately. Examples of gene expression profiling studies on breast cancer prognosis include [3], which used Affymetrix U133A microarrays and identified 97 genes including UBE2C, KPNA2, TPX2, FOXM1, STK6, CCNA2, BIRC5, and MYBL2. Ivshina et al. [4] reported similar findings from a concurrent, independent study. Researchers at the Netherlands Cancer Institute identified a 70-gene prognostic signature [5]. Many genes involving the hallmarks of cancer were included: cell cycle, metastasis, angiogenesis, and invasion. This gene signature was then validated on an independent cohort of 295 patients [6]. References to more studies can be found in [1,2]. When searching for genomic measurements with predictive power for breast cancer prognosis, it is necessary to account for the inherent coordination among genes. Such coordination can be described with the pathway structure, where pathways are composed of multiple genes with coordinated biological functions. In cancer genomic studies, tremendous effort has been devoted to pathway based analysis. "Pathway analysis is a promising tool to identify the mechanisms that underlie diseases, adaptive physiological compensatory responses, and new avenues for investigation" [7]. Compared with individual gene based analysis, pathway based analysis may lead to results that are more reproducible and more interpretable. Examples of pathway analysis methods include the gene set enrichment analysis (GSEA) [8], the Globaltest approach [9], the Maxmean approach [10], and others. We refer to [11-13] for comprehensive reviews on the subject. Consider a pathway composed of m genes. Denote X = (X1, ..., Xm)' as the gene expressions. Consider breast cancer relapse-free survival. We refer to the "Methods" section for detailed descriptions of the data and model setup. Determining the predictive power amounts to determining whether there exists a length-m vector β such that β'X can be used to separate patients into groups with different survival risks. We first note that, (a) Different pathways have different biological functions. Thus, it is reasonable to study each pathway separately. Among the many pathways, only a few have predictive power for cancer development. Among genes within predictive pathways, there are a subset having small to moderate predictive power, whereas the remainder are "noisy" genes. Within each pathway, instead of investigating each Xi separately (i.e, the marginal effect of each gene), it is more sensible to study β'X (i.e, the joint effects of multiple genes); (b) Cancer genomic studies often have small sample sizes, and sizes of gene pathways can be large. When investigating the joint effects of multiple genes in a pathway, if the same dataset is used for estimation of β as well as evaluation of predictive power, the evaluation can be seriously biased [14]. Ideally, there should be two independent datasets: a training set and a testing set. β should be generated using only subjects in the training set. Then predictions can be made for subjects in the testing set using the training set estimate, and the predictive power can be evaluated. Although there are many existing pathway analysis methods, they are not suitable for detecting predictive gene pathways for one or more of the following reasons. (a) For a specific pathway, they analyze each gene separately, and then draw conclusions on the pathway by combining results on individual genes. Such methods, including the GSEA and Maxmean, are suitable for answering "which pathways are enriched with genes that are marginally differentially expressed". They cannot quantify the joint effects of genes in a pathway; (b) They focus on the model fitting aspect of genes, as opposed to prediction. When studying one or a small number of genes, model fitting performance can be a reasonable proxy for prediction performance. However, when investigating a moderate to large number of genes, because of the possibility of overfitting, model fitting performance can be a biased proxy for prediction; and (c) They analyze only a single dataset. Cancer genomic studies have small sample sizes and a large number of gene expressions. Results obtained from analysis of a single dataset may lack reliability [15]. In this article, we propose a new method for detection of predictive gene pathways. It has the following desirable features. (a) For each pathway, it uses a single statistical model to describe the effects of all genes in the pathway. Thus, it can account for the joint effects of genes; (b) A penalized approach is used to construct β. The penalized approach can carry out regularized estimation and gene selection simultaneously. Adopting the penalized approach has been motivated by the following considerations. First, when the pathway sizes are larger than or comparable to the sample size, the penalized approach can effectively avoid overfitting. Second, even in a predictive pathway, there may still exist noisy genes. The penalized approach can separate predictive genes from noisy ones and use only predictive genes in the statistical models. This can lead to better performance than using all the genes; (c) A random partition is used to split data into

Read the complete text