In the last years, several proposals that make use of text-mining methods in the context of microarrays have been made such as GEISHA (Oliveros et al., 2000), MedMiner (Tanabe et al., 1999), ConceptMaker (Kuffner et al., 2005) or others (Krallinger and Valencia, 2005). Nevertheless, although such programs provide biological terms related to the query gene(s), they do not implement a robust statistical framework to assess the significance of the results found beyond simple measurements of enrichment. And especially, there is nothing like the functional profiling method presented in MarmiteScan that, similarly to FatiScan (Al-Shahrour et al., 2005), directly tests the behavior of blocks of functionally related genes, and does not require of a previous step of gene selection. It is worth mentioning that the segmentation test implemented in this tool does not depend on the original data for obtaining P-values, but only on the gene ranking. As a result, many different experimental designs (two-class comparisons, survival, correlation to any parameter, etc.) can be tested providing these produce a gene ranking.
To our knowledge, Marmite and MarmiteScan are the only applications in which functional profiling, based on text-mining, is performed in user-friendly environment within the proper statistical framework.
Answers to all your Biology Questions
