Costa 1, Alexander Schnhuth 2 and Alexander Schliep 1, 1 Max Planck Institute for Molecular Genetics, Department of Computational Molecular Biology, Ihnestrasse 73, 14195 Berlin, Germany and 2 Center for Applied Computer Science, University of Cologne, Weyertal 80, 50937 Cologne, Germany Received on January 05, 2004; accepted on February 7, 2005 Advance Access publication February 8, 2005 ABSTRACT Summary: The Graphical Query Language (GQL) is a set of tool s for the analysis of gene expression time - courses.

They allow a user to pre-process the data, to query it for interesting patterns, to per for m model-based clustering or mixture estimation, to include subsequent refinements of clusters and, finally, to use other biological resources to evaluate the results.

Analyses are carried out in a graphical and interactive environment, allowing expert intervention in all stages of the data analysis.

As the knowledge discovery process in the analysis of biological data is human-centric, a high degree of interactivity is an important characteristic of the Graphical Query Language (GQL).

What we have striven for is a set of application tool s which lets a user visualize and analyze time -course data interactively, evaluate hypotheses about the data and compare the results with other sources of biological data.

GQL allows to integrate prior knowledge and it maintains a high degree of robustness with respect to noise and missing data, in order to arrive at unambiguous groups of time - courses.

The main contributions of our method is the use of linear hidden Markov models (HMMs) to represent groups of gene s showing the same qualitative behavior and their combination into a classical mixture model; this has been shown to be an intuitive and meaningful choice (Schliep et al., 2003, 2004).

SOFTWARE DESCRIPTION GQL is divided into two main applications: GQLQuery and GQL- Cluster.

GQLQuery allows the user to either create a new HMM or load an existing one in order to query a set of time - courses for interesting temporal patterns (Fig.

Modifications of the models parameters in the tool interface are simultaneously reflected in the time - courses panel.

By changing the similarity rank threshold, the user can control the stringency of the query, and thus select only To whom correspondence should be addressed.

The modified models and query results can be saved for later analyses.

GQLCluster implements the methods for estimating clusterings or mixtures of time - courses and for the post- analysis of the results.

As a first step, the time - courses can be filtered to exclude non-expressed gene s.

GQLCluster provides an n-fold filter and a non-constant filter.

There is no need to do any pre-processing concerning missing data, since the estimation methods can deal internally with these values.

After filtering, different estimation procedures can be applied to find interesting groups of gene s in the data.

As model-based estimation procedures require the provision of an initial collection of models, GQLCluster has implemented three easy-to-use and welljustified methods for this.

They can be either defined by the user, e.g.

GQLQuery, randomly gene rated or estimated from the input data.

Subsequent to the creation of an initial model set, three types of estimations are applicable 2544 The Author 2005.

For Permissions, please email: journals.permissionsoupjournals.org Downloaded from for matics.ox for djournals.org by guest on November 20, 2012 Page 2: Fig.

After estimation, GQLCLuste.

Bioinformatics Software For Mac