Given the score methods described above, genes can be ranked based on score if experiment classes are defined. J-Express include several variants for finding good marker genes, either by ranking gene by gene or by looking at combinations (pairs) of genes.
Individual ranking
This ranking method computes a score for each gene profile, and ranks the list of genes by score. The genes with highest (absolute) score are reported on top of the list.
Greedy pairs
The greedy pairs ranking method first ranks all genes by individual ranking. Subsequently the highest scoring gene gi is paired with the gene gj that gives the highest gene pair score. The gene pair score is computed by projecting the expression values of the two genes onto the diagonal linear discriminant axis, and then taking the score of the transformed data points. After the first pair has been selected, the highest ranked gene remaining gs is paired with the gene gr that maximizes the pair score, and so on. See Bø et al. [2] for further details.
All pairs
Unlike greedy pairs, this method examines all possible gene pairs by computing the pair score for all pairs. The pairs are then ranked by pair score, and the gene ranking list is compiled by selecting nonoverlapping pairs, and selecting highest scoring pairs first. This method is computationally intensive, and may take a while to terminate. See Bø et al. [2] for further details.
[1] Bhattacharyya GK and Johnson RA: Statistical concepts and methods. Wiley, 1977.
[2] Bø TH and Jonassen I: New feature subset selection procedures for classification of expression profiles. Genome Biology, 3(4):research0017.1-0017.11, 2002. Available online: http://genomebiology.com/2002/3/4/research/0017.1.
[3] Dudoit S, Fridlyand J, Speed T: Comparison of discrimination methods for the classification of tumors using gene expression data. Technical report no. 576, Department of Statistics, University of California, 2000.
[4] Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeeck M, Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA, et al.: Molecular classi_cation of cancer: Class discovery and class prediction by gene expression monitoring. Science 1999, 286:531-537.
Please refer to the following paper for method description:
New feature subset selection procedures for classification of expression
profiles
Trond Hellem Bø , Inge Jonassen
Department of Informatics, University of Bergen, N-5020 Bergen, Norway
Genome Biology 2002 3(4): research0017.1-0017.11