"How to integrate the huge volume of disparate data - on gene expression, protein interactions and the vast output of literature both inside and outside laboratories - to find out what is important", stated Dr. Michael Schroeder, Professor of the Bioinformatics group at Dresden Technical University and co-ordinator of this IST-funded project. "I attended a workshop recently, held by the W3 consortium, and many of the companies there said that this was the biggest problem they face."
Currently, pharmaceutical and biotech companies produce vast quantities of raw data on the problems that interest them. Micro-arrays process thousands of samples to discover what genes are over-expressing. These over-expressing genes - numbering sometimes in their thousands, too - create proteins. The researchers then need to discover what protein interactions are taking place among all the different proteins created by the over-expressing genes. This is not trivial.
If a researcher can identify protein interactions they then need to do a search on their company intranet to see what other work company labs have produced relevant to the topic. Since the data output at biotech labs is vast, this is also not trivial.
Finally, the researcher must perform a search of academic journals to find relevant journal papers. Currently PubMed, the most important public literature database available, has 15.000.000 entries, and the number is growing every day. Finding relevant data there is again not a trivial task.
Dr. Schroeder gave an example. "The medical faculty here were studying pancreatic tumours. They found 1000 genes over-expressing. Using our software they were able to find, among others, three protein interactions that were particularly relevant. Using our literature search ontology they were able to discover that two of these interactions were novel. They are now going to study these novel interactions more closely", he stated.
This is how the project will help companies integrate all the data they need to make relevant discoveries using a BioGrid. A BioGrid is essentially a data and computational Grid created through a suite of tools developed by the project. Biotech and life science companies will use these tools internally to integrate data. They can then also use the same tools to create ad-hoc distributed processing networks to crank through computationally-intense problems, if needed.
Here's how it works. One element of the software suite analyses over-expressing genes discovered during micro assays to establish what proteins become encoded. This uses standard techniques. A second analysis tool in the suite predicts what possible protein-protein interactions are taking place. This is novel. When a gene encodes a protein, the protein folds up into a unique shape, forming a 3D structure. This structure can only interact, or fit, with some proteins, but not others, like pieces of a jigsaw puzzle.
BioGrid's protein interaction software includes a database of the 20.000 known protein structures and uses that database to identify which ones could potentially interact, among the thousands of proteins created by the over-expressing genes. This focuses research efforts on the most interesting candidates for a particular problem, like, for example, pancreatic tumours.
Once interesting potential protein interactions are known, BioGrid's ontology-based search technology can mine company or journal data for any relevant information. The Gene Ontology (GO) was established by the life sciences as a vocabulary to describe all the different genetic processes. There are 20.000 terms in this vocabulary so far.
"The problem is researchers never use exact terms from the Gene Ontology to do their search of the literature. Our innovation was to create an algorithm that intelligently matches the terms used by researchers to the Gene Ontology", Dr. Schroeder stated.
This smart search produces a list of all the possible GO terms and lists all the articles relevant to that specific term. Researchers are presented with a vast quantity of information broken down by sub-topic, so they can quickly drill down to the most relevant information.
Linking all these software tools together is a rules-based Java scripting language called Prova, also developed by the BioGrid team. It is the glue that sticks the Gene Expression, Protein Interaction and ontology-based literature analysis together into an integrated, cohesive unit. "It's an open source language, available at http://www.prova.ws, and about 20 groups are using it around the world right now. We made it open source because you need to develop a community to keep a programming language alive", stated Dr. Schroeder.
So biotech companies can establish their own BioGrids on their intranet, some of the researchers involved in BioGrid created the spin-off company Molgenis supported by a BioPartner first stage Grant. In addition other partners are in the process of obtaining seed funding for a spin-off company called Transinsight.
"BioGrid is a knowledge Grid that offers the proper software and support so that companies can do rapid prototyping for their data-integration and large computations", stated Dr. Schroeder. "Transinsight will continue working on new, innovative and novel solutions to data problems faced by the life sciences industry", hoped Dr. Schroeder. "The company will look at natural language processing and the semantic Grid for the life sciences. It is currently running two feasibility studies with two international customers. Molgenis provides software to support the storage and analysis of gene expression data."
Many of the tools developed by BioGrid are available for public use. The ontology-based search is available at GoPubMed.org, while the protein interaction database is at Scoppi.org. But GoPubMed is limited to 100 searches while Scoppi does not include the predictive analysis tool. These limitations will not be there for commercial clients. For those, BioGrid offers an earthmover for data mining.
For more information you can contact Michael Schroeder, Professor in Bioinformatics at the Biotec/Dept. of Computing, TU Dresden, Tatzberg 47-51, D-01307 Dresden, Germany, Tel.: +49-351-46340060, Fax: +49-351-46340061. This article was reprinted from the IST Results Web site.