DFP: a Bioconductor package for fuzzy profile identification and gene reduction of microarray data

• Main Authors: Fdez-Riverola Florentino, Díaz Fernando, Álvarez Rodrigo, Glez-Peña Daniel
• Format: Article
• Language: English
• Published: BMC 2009-01-01
• Series: BMC Bioinformatics
• Online Access: http://www.biomedcentral.com/1471-2105/10/37

<p>Abstract</p> <p>Background</p> <p>Expression profiling assays done by using DNA microarray technology generate enormous data sets that are not amenable to simple analysis. The greatest challenge in maximizing the use of this huge amount of data is to develop algorithms to interpret and interconnect results from different genes under different conditions. In this context, fuzzy logic can provide a systematic and unbiased way to both (<it>i</it>) find biologically significant insights relating to meaningful genes, thereby removing the need for expert knowledge in preliminary steps of microarray data analyses and (<it>ii</it>) reduce the cost and complexity of later applied machine learning techniques being able to achieve interpretable models.</p> <p>Results</p> <p>DFP is a new Bioconductor R package that implements a method for discretizing and selecting differentially expressed genes based on the application of fuzzy logic. DFP takes advantage of fuzzy membership functions to assign linguistic labels to gene expression levels. The technique builds a reduced set of relevant genes (FP, <it>Fuzzy Pattern</it>) able to summarize and represent each underlying class (pathology). A last step constructs a biased set of genes (DFP, <it>Discriminant Fuzzy Pattern</it>) by intersecting existing fuzzy patterns in order to detect discriminative elements. In addition, the software provides new functions and visualisation tools that summarize achieved results and aid in the interpretation of differentially expressed genes from multiple microarray experiments.</p> <p>Conclusion</p> <p>DFP integrates with other packages of the Bioconductor project, uses common data structures and is accompanied by ample documentation. It has the advantage that its parameters are highly configurable, facilitating the discovery of biologically relevant connections between sets of genes belonging to different pathologies. This information makes it possible to automatically filter irrelevant genes thereby reducing the large volume of data supplied by microarray experiments. Based on these contributions <smcaps>GENE</smcaps>CBR, a successful tool for cancer diagnosis using microarray datasets, has recently been released.</p>