SAMPL - Sampling And Model-Selection PipeLine : Systematically Combining Stable Gene Selection and Classification
Two goals of gene expression analysis are classification and gene selection. Combinations of gene selectors with classifiers were evaluated on random subsets of arrays. Consensus gene signatures are constructed from genes frequently selected. Model reliability is estimated by the variance of the classification performances. The method finds reliable models, even when classification performance is variable. Manual literature screening verified the relevance of the genes identified from an osteoarthritis dataset.