Jennifer Pittman

Gene expression predictors of breast cancer outcomes

BACKGROUND Correlation of risk factors with genomic data promises to provide specific treatment for individual patients, and needs interpretation of complex, multivariate patterns in gene expression data, as well as assessment of their ability to improve clinical predictions. We aimed to predict nodal metastatic states and relapse for breast cancer patients. METHODS We analysed DNA microarray data from samples of primary breast tumours, using non-linear statistical analyses to assess multiple patterns of interactions of groups of genes that have predictive value for the individual patient, with respect to lymph node metastasis and cancer recurrence. FINDINGS We identified aggregate patterns of gene expression (metagenes) that associate with lymph node status and recurrence, and that are capable of predicting outcomes in individual patients with about 90% accuracy. The metagenes defined distinct groups of genes, suggesting different biological processes underlying these two characteristics of breast cancer. Initial external validation came from similarly accurate predictions of nodal status of a small sample in a distinct population. INTERPRETATION Multiple aggregate measures of profiles of gene expression define valuable predictive associations with lymph node metastasis and disease recurrence for individual patients. Gene expression data have the potential to aid accurate, individualised, prognosis. Importantly, these data are assessed in terms of precise numerical predictions, with ranges of probabilities of outcome. Precise and statistically valid assessments of risks specific for patients, will ultimately be of most value to clinicians faced with treatment decisions.

Patterns of Gene Expression That Characterize Long-term Survival in Advanced Stage Serous Ovarian Cancers

Purpose: A better understanding of the underlying biology of invasive serous ovarian cancer is critical for the development of early detection strategies and new therapeutics. The objective of this study was to define gene expression patterns associated with favorable survival. Experimental Design: RNA from 65 serous ovarian cancers was analyzed using Affymetrix U133A microarrays. This included 54 stage III/IV cases (30 short-term survivors who lived <3 years and 24 long-term survivors who lived >7 years) and 11 stage I/II cases. Genes were screened on the basis of their level of and variability in expression, leaving 7,821 for use in developing a predictive model for survival. A composite predictive model was developed that combines Bayesian classification tree and multivariate discriminant models. Leave-one-out cross-validation was used to select and evaluate models. Results: Patterns of genes were identified that distinguish short-term and long-term ovarian cancer survivors. The expression model developed for advanced stage disease classified all 11 early-stage ovarian cancers as long-term survivors. The MAL gene, which has been shown to confer resistance to cancer therapy, was most highly overexpressed in short-term survivors (3-fold compared with long-term survivors, and 29-fold compared with early-stage cases). These results suggest that gene expression patterns underlie differences in outcome, and an examination of the genes that provide this discrimination reveals that many are implicated in processes that define the malignant phenotype. Conclusions: Differences in survival of advanced ovarian cancers are reflected by distinct patterns of gene expression. This biological distinction is further emphasized by the finding that early-stage cancers share expression patterns with the advanced stage long-term survivors, suggesting a shared favorable biology.

Gene Expression Phenotypes of Atherosclerosis

Objective—Fulfilling the promise of personalized medicine by developing individualized diagnostic and therapeutic strategies for atherosclerosis will depend on a detailed understanding of the genes and gene variants that contribute to disease susceptibility and progression. To that end, our group has developed a nonbiased approach congruent with the multigenic concept of complex diseases by identifying gene expression patterns highly associated with disease states in human target tissues. Methods and Results—We have analyzed a collection of human aorta samples with varying degrees of atherosclerosis to identify gene expression patterns that predict a disease state or potential susceptibility. We find gene expression signatures that relate to each of these disease measures and are reliable and robust in predicting the classification for new samples with >93% in each analysis. The genes that provide the predictive power include many previously suspected to play a role in atherosclerosis and additional genes without prior association with atherosclerosis. Conclusion—Hence, we are reporting a novel method for generating a molecular phenotype of disease and then identifying genes whose discriminatory capability strongly implicates their potential roles in human atherosclerosis.

Genomic Prediction of Locoregional Recurrence After Mastectomy in Breast Cancer

PURPOSE This study aims to explore gene expression profiles that are associated with locoregional (LR) recurrence in breast cancer after mastectomy. PATIENTS AND METHODS A total of 94 breast cancer patients who underwent mastectomy between 1990 and 2001 and had DNA microarray study on the primary tumor tissues were chosen for this study. Eligible patient should have no evidence of LR recurrence without postmastectomy radiotherapy (PMRT) after a minimum of 3-year follow-up (n = 67) and any LR recurrence (n = 27). They were randomly split into training and validation sets. Statistical classification tree analysis and proportional hazards models were developed to identify and validate gene expression profiles that relate to LR recurrence. RESULTS Our study demonstrates two sets of gene expression profiles (one with 258 genes and the other 34 genes) to be of predictive value with respect to LR recurrence. The overall accuracy of the prediction tree model in validation sets is estimated 75% to 78%. Of patients in validation data set, the 3-year LR control rate with predictive index more than 0.8 derived from 34-gene prediction models is 91%, and predictive index 0.8 or less is 40% (P = .008). Multivariate analysis of all patients reveals that estrogen receptor and genomic predictive index are independent prognostic factors that affect LR control. CONCLUSION Using gene expression profiles to develop prediction tree models effectively identifies breast cancer patients who are at higher risk for LR recurrence. This gene expression-based predictive index can be used to select patients for PMRT.

Adaptive Splines and Genetic Algorithms

Most existing algorithms for fitting adaptive splines are based on nonlinear optimization and/or stepwise selection. Stepwise knot selection, although computationally fast, is necessarily suboptimal while determining the best model over the space of adaptive knot splines is a very poorly behaved nonlinear optimization problem. A possible alternative is to use a genetic algorithm to perform knot selection. An adaptive modeling technique referred to as adaptive genetic splines (AGS) is introduced which combines the optimization power of a genetic algorithm with the flexibility of polynomial splines. Preliminary simulation results comparing the performance of AGS to those of existing methods such as HAS, SUREshrink and automatic Bayesian curve fitting are discussed. A real data example involving the application of these methods to a fMRI dataset is presented.

Fitting optimal piecewise linear functions using genetic algorithms

Constructing a model for data in R/sup 2/ is a common problem in many scientific fields, including pattern recognition, computer vision, and applied mathematics. Often little is known about the process which generated the data or its statistical properties. For example, in fitting a piecewise linear model, the number of pieces, as well as the knot locations, may be unknown. Hence, the method used to build the statistical model should have few assumptions, yet, still provide a model that is optimal in some sense. Such methods can be designed through the use of genetic algorithms. We examine the use of genetic algorithms to fit piecewise linear functions to data in R/sup 2/. The number of pieces, the location of the knots, and the underlying distribution of the data are assumed to be unknown. We discuss existing methods which attempt to solve this problem and introduce a new method which employs genetic algorithms to optimize the number and location of the pieces. Experimental results are presented which demonstrate the performance of our method and compare it to the performance of several existing methods, We conclude that our method represents a valuable tool for fitting both robust and nonrobust piecewise linear functions.

Information conversion, effective samples, and parameter size

Consider the relative entropy between a posterior density for a parameter given a sample and a second posterior density for the same parameter, based on a different model and a different data set. Then the relative entropy can be minimized over the second sample to get a virtual sample that would make the second posterior as close as possible to the first in an informational sense. If the first posterior is based on a dependent dataset and the second posterior uses an independence model, the effective inferential power of the dependent sample is transferred into the independent sample by the optimization. Examples of this optimization are presented for models with nuisance parameters, finite mixture models, and models for correlated data. Our approach is also used to choose the effective parameter size in a Bayesian hierarchical model.

The construction and assessment of a statistical model for the prediction of protein assay data.

The focus of this work is the development of a statistical model for a bioinformatics database whose distinctive structure makes model assessment an interesting and challenging problem. The key components of the statistical methodology, including a fast approximation to the singular value decomposition and the use of adaptive spline modeling and tree-based methods, are described, and preliminary results are presented. These results are shown to compare favorably to selected results achieved using comparitive methods. An attempt to determine the predictive ability of the model through the use of cross-validation experiments is discussed. In conclusion a synopsis of the results of these experiments and their implications for the analysis of bioinformatic databases in general is presented.

Multilayer perceptrons and fractals

Abstract In this article, a mathematical relationship between the gradient descent technique and contractive maps is examined. This relationship is based upon the observation that the convergence of the gradient descent technique can be proved using results in fractal theory — more specifically, results concerning contractive maps — as opposed to results based on Taylors series. This proof, involving the eigenvalues of the Hessian matrix of the gradient descent techniques objective function, is presented. A simple example is given in which steps from the aforementioned proof are used to find conditions under which a specific multilayer perceptron is guaranteed to converge. Since the gradient descent technique is used in multilayer perceptrons, and contractive maps give rise to fractals, a theoretical relationship is thus established between multilayer perceptrons and fractals.

Gene expression profiles that predict response to platinum and identify patterns of pathway deregulation in advanced ovarian cancer

5031 Background: Platinum drugs are the most active agents in epithelial ovarian cancer, however 30% of patients with advanced disease have an incomplete response. Current treatment strategies result in many patients receiving multiple cycles of platinum-based chemotherapy without complete response. We have developed gene expression profiles that predict response to platinum therapy, with the aim of individualizing therapy. In parallel, we identified patterns of oncogenic pathway deregulation within the non-responding patient group that provides a rational basis for directing targeted therapeutics. Methods: Affymetrix GeneChips were used to measure gene expression in 120 stage III/IV serous ovarian cancers from patients who underwent primary surgical cytoreduction followed by platinum therapy. Expression data was compared between cancers that demonstrated a complete and incomplete response to platinum therapy. Predictive models were developed to distinguish patients likely to respond to therapy. In additi...