Servicio de Información Comunitario sobre Investigación y Desarrollo - CORDIS

New algorithms developed and implemented in data analysis tool Expression Profiler

New algorithms for predicting gene function, for identifying periodic genes in micro-array data, for identifying co-expressed blocks of genes and samples, new data normalisation algorithms, and a novel algorithm for cluster comparison have been developed and published in high impact journals.

In particular new methods have been developed for predicting gene functions from micro-array data, for identifying periodic genes in micro-array data, K-medoids clustering algorithm and for comparing results of different clusterings. Jointly with a wide range of collaborators, a numerous existing algorithms have been modified and implemented in Expression Profiler. These include the "signature" algorithm (developed in Weizman institute), "in between group analysis" algorithm (developed in Cork University), and various normalisation algorithms. All are available via Expression Profiler interface (see

Hierarchical and non-hierarchical clustering are among of the most widely used gene expression data analysis methods. Different clustering methods and different parameters often produce different results. Understanding how different clustering results relate to each other is important if we are to understand the biological relevance of different clusters and the underlying data. The clustering results can sometimes be rather different, in which case the problem often goes beyond one-to one relationship between clusters on in each clustering. We developed a clustering comparison method and its implementation that finds the correspondence between groups of clusters in two different clustering results.

The number of clusters in the results to be compared may be very different, and we allow the comparison of either two non-hierarchical clusterings (such as K-means), or a flat and a hierarchical clustering. Using simulated data we show that our method can be used to approximate the "true" clusters, while using real gene expression data we show that we can restore biologically meaningful clusters. The method is available online as a part of tool-set Expression Prifiler The software is open source

Reported by

United Kingdom
See on map
Síganos en: RSS Facebook Twitter YouTube Gestionado por la Oficina de Publicaciones de la UE Arriba