Catálogo de publicaciones - libros

Lung cancer is the leading cause of cancer death. Gene expression profiling of lung cancer may provide one method to increase our understanding of this very heterogeneous disease and potentially identify new approaches for early diagnosis, prognosis or treatment. This brief review examines some of the issues that are associated with the clinical presentation of this disease and some important questions that might be addressed using gene expression profiling experiments when applied to human lung cancer.

Finding associations between expression profiles and simple phenotypic data such as class labels has been studied extensively, including prediction algorithms for new samples based on these relationships. However, much work is needed to link expression profiles to more complex response variables, most notably survival data with censoring. Reducing the survival data to a short-term versus long-term survival indicator or using survival curves merely to demonstrate the difference between clusters of samples is not an efficient use of the data. We review some of the progress and challenges in this area. We discuss the need for more consistent results among studies done on different microarray platforms, for development of sample-specific predictive scoring schemes, and for a more comprehensive analysis that incorporates other prognostic factors and clearly demonstrates the added value of expression profiling over current protocols.

Microarray technology provides exciting tools for monitoring expression levels of hundreds or thousands of genes simultaneously. Good microarray studies have clear objectives. To make meaningful statistical interpretation of study results obtained from microarray experiments, the design of the experiments must consider some degree of replication to allow for the description of sources of variations. In this article, we present an overview of replicate designs that incorporate measurement variability to address its study objectives.

Our goal in this work was to pool information across microarray studies conducted at different institutions using two different versions of Affymetrix chips to identify genes whose expression levels offer information on lung cancer patients’ survival above and beyond the information provided by readily available clinical covariates. We combined information across chip types by identifying “matching probes” present on both chips, and then assembling them into new probesets based on Unigene clusters. This method yielded comparable expression level quantifications across chips without sacrificing much precision or significantly altering the relative ordering of the samples. We fit a series of multivariable Cox models containing clinical covariates and genes and identified 26 genes that provided information on survival after adjusting for the clinical covariates, while controlling the false discovery rate at 0.20 using the Beta-Uniform mixture method. Many of these genes appeared to be biologically interesting and worthy of future investigation. Only one gene in our list has been mentioned in previously published analyses of these data. It appears that the increased statistical power provided by the pooling was key in finding these new genes, since only nine out of the 26 genes were detected when we apply these methods to the two data sets separately, i.e., without pooling.

The application of gene expression microarray technology has the potential to have a large impact in the area of oncology. There is a need to be able to identify genes associated with prolonged or reduced survival, to aid decisions regarding patient treatment and care. In addition these genes can be targeted in drug research to aid discovery and development of novel treatments. This paper uses two published Affymetrix datasets and combines the information from adenocarcinoma lung tumors to identify genes associated with survival. Kaplan-Meier survival analysis, Cox proportional hazards models and analysis of variance are used for the data analyses. The results are combined across the two datasets using Fisher’s chi-squared meta-analysis based on p-value aggregation. The false discovery rate (FDR) adjustment is made to the final pvalues.

High throughput technologies such as microarray, mass spectrometry and nuclear magnetic resonance, have generated large volumes of valuable data for biology research. Researchers often face the challenges of integrating data from different sources and of identifying potential biomarkers that are highly associated with disease, drug safety, and efficacy. We present several solutions to these challenges through two Affymetrix microarray studies aimed at providing new insights into lung cancer biology. The Harvard dataset and the Michigan dataset were integrated to identify genes that were predictive of cancer survival. Quantile normalization of expression measures was applied to make the two datasets comparable. Genes highly associated with survival were identified and survival tree analysis on the combined data was performed to predict mortality. The candidate genes could be useful for lung cancer disease prediction and cancer therapy. The methodologies for integration and analysis of multiple gene expression data have been shown to perform well and could be generalized to broader applications.

In order to comprehensively identify genes with expression levels that correlate with survival for patients with lung adenocarcinoma, we combined data across the Harvard and Michigan studies. Two different versions of Affymetrix oligonucleotide microarrays were used in these two studies. We proposed combining arrays of different platforms by assigning weights to the expression levels of each gene across data sets based on the entropy of the residual matrix. In each data set, the expression level of each gene is quantified by the “reduced” model proposed by Li and Wong [2001], which is equivalent to a method using the singular value decomposition. We combined information across different chip types by first identifying common genes on the two chip types, and then assigning weights based on residual entropy for each gene. To incorporate clinical information, especially survival data, in detecting important genes, we proposed a new method based on weighted t-tests (wtt). The survival information can be absorbed into a set of weights assigned to the expression intensities across all the arrays or subjects, based on the predicted median survival time using the Cox proportional hazards model. Important genes can be identified by comparing the survival-weighted t-tests with another t-test comparing the cancer patients to the reference group, and error rates can be controlled by permutation procedures.

In many microarray studies the primary objective is to identify, from a large panel of genes, those which are prognostic markers of a censored survival endpoint such as time to disease recurrence or death. These genes are considered prognostic in that their respective expressions are associated, in an appropriate sense, with the survival endpoint of interest. From a practical point of view, this requires not only specifying a appropriate measure of association and a suitable statistic thereof, but also, as the number of genes is large, proper handling of the consequential issue of multiplicity. In this paper, we will address the aforementioned issues by utilizing a general correlation measure and a non-parametric statistic, and by controlling the family-wise error rate by employing permutation resampling. Comprehensive simulation studies are conducted to investigate the statistical properties of the proposed procedure. The proposed procedure is demonstrated with microarray data.

We propose a simple data analysis procedure that aims to uncover an association between gene expression and the status of a clinical outcome variable. Rather than focus on differences in group means, as is usually done, we search for pairs of genes such that the strength or direction of their association is linked to the value of the outcome variable. This more complex pattern of gene expression, which we call “differential correlation”, may be especially relevant in studying clinical outcomes such as survival and grade, since it has often been difficult to identify marker genes whose mean expression varies directly with such outcomes. In applying our method to two lung cancer microarray data sets, we discovered that a substantially greater number of genes are likely to be associated with clinical outcomes such as tumor stage via differential correlation than are associated via changes in mean expression.