Improving Microarray Sample Size Using Bootstrap Data Combination

Abstract

Microarray technology has enabled us to simultaneously measure the expression of thousands of genes. Using this high-throughput technology, we can examine subtle genetic changes between biological samples and build predictive models for clinical applications. Although microarrays have dramatically increased the rate of data collection, sample size is still a major issue when selecting features. Previous methods show that combining multiple microarray datasets improves feature selection using simple methods such as fold change. We propose a wrapper-based gene selection technique that combines bootstrap estimated classification errors for individual genes across multiple datasets and reduces the contribution of datasets with high variance. We use the bootstrap because it is an unbiased estimator of classification error that is also effective for small sample data. Coupled with data combination across multiple datasets, we show that our meta-analytic approach improves the biological relevance of gene selection using prostate and renal cancer microarray data.

Quality control of highly multiplexed proteomic immunostaining with quantum dots: Correcting for crosstalk

“…Most reputable biomarker discovery work now includes some form of targeted validation, either through quantitative Polymerase Chain Reaction (PCR) or independent microarray experimentation. In addition to quality assurance through validation, methods also exist to proactively increase quality by combining heterogeneous datasets or incorporating existing knowledge such as known biomarkers or gene ontology information [6], [7]. …”