Improved bolstering error estimation for gene ranking.

Many methods have been proposed to identify differentially expressed genes in diseased tissues. The performance of the method is closely related to the evaluation metric. We examine several error estimation algorithms (i.e., cross validation, bootstrap, resubstitution, and resubstitution with bolstering) for three classifiers (i.e., support vector machine, Fisher's discriminant, and signed distance function). To control the classifier's data-overfitting problem, usually caused by small sample size for many real datasets, we generate synthetic datasets based on real data. This way, we can monitor sample size impact when evaluating the metrics. We find that resubstitution with bolstering has the best result, especially with respect to computational efficiency. However, classical bolstering tends to bias in high dimensions. Thus, we further investigate ways to reduce bolstering estimation bias without increasing computational intensity. Results of our investigation indicate that the estimator tends to become unbiased as the sample size increases. We also find that modified bolstering is the best among all metrics in terms of estimation accuracy and computational efficiency.

Reproducibility of differential gene detection across multiple microarray studies.

Although expression profiling of various diseases to identify interesting genes is a well-established methodology, it still faces many challenges. Labs often have difficulty reproducing results on different microarray platforms. Microarray manufacturers use different clones to represent similar genes on various platforms. Consequently, researchers struggle to integrate data published in literature and databases. Even results from identical microarray platforms may not correlate due to technical variability between labs. We seek some degree of congruity between the same microarray platforms implemented at multiple test sites. We analyze two prostate cancer datasets from commercially synthesized oligonucleotide arrays (Affymetrix HG-U95v2). Our analysis focuses on determining reproducibility in identifying differentially expressed genes using fold change and t-tests. We use p-values to compare specificity and sensitivity of the methods applied to each dataset. Findings indicate that, even though both datasets use the same microarray platform, differences in experimental design and test conditions result in variations when detecting differentially expressed genes.