Researchers from the Fraunhofer Institute for Algorithms and Scientific Computing (SCAI; Saint Augustin, Germany) and the Jülich Supercomputing Center of Forschungszentrum Jülich (JSC; Germany) have used their considerable grid computing infrastructures for a new application in scientific computing: the large-scale annotation of biomedical and chemical texts and images in pharmaceutical patents. This will allow patent searches of an unparalleled power. Now, queries provide interesting insights into intersections between biology and chemistry, and the analysis of chemistry is truly multimodal in the sense that text- and image-based information can be analyzed simultaneously.
More than 50,000 patents describing inventions in pharmaceutical chemistry have been processed on the large-scale computing grid infrastructures at SCAI and JSC. Automated “named entity recognition” services have identified and annotated: biologic entities in text (e.g., protein names; gene names; gene polymorphisms; cell types); medical entities in text (e.g., disease names; pathology terms; risk factor terminology); as well as chemical information in text (e.g., drug names; expressions following the naming standards of the International Union of Pure and Applied Chemistry [IUPAC]); and images (e.g., chemical structure depictions).
The grid middleware UNICORE (Uniform Interface to Computing Resources) was used to manage the annotation services in the grid infrastructure, to control the streams of input and output data from the patents database to the annotation services, and to monitor the overall progress.
“This large-scale experiment opens new perspectives in scientific computing,” commented Prof. Dr. Martin Hofmann-Apitius, head of the department of bioinformatics at Fraunhofer SCAI. “This type of application goes way beyond the usual simulation applications that we are used to in the scientific computing community.”
Up to now, text-mining applications have only been run on bibliographic databases of life sciences and biomedical information such as MEDLINE. But the extension towards a multimodal analysis including annotation of text- and image-based information in full text documents on grid infrastructures has never been done before.
“We are pleased to see that our institute, which has a strong record in numerical simulation, has contributed to a new field of applications for supercomputers: what we call knowledge computing is likely to become a new discipline on its own,” emphasized Prof. Dr. Ulrich Trottenberg, director of Fraunhofer SCAI.
“UNICORE made it possible to run this experiment at such a large scale in computing grid infrastructures at SCAI and JSC,” stated Dr. Achim Streit, head of Distributed Systems and Grid Computing at JSC. “The powerful workflow and data management capabilities of UNICORE allowed to annotate the patents in a seamless and automated way. A supercomputer connected by UNICORE to the infrastructure of the German Grid Initiative [D-Grid] was used to perform the knowledge extraction. This initial step of the experiment demonstrates what is possible today and shows the potential for more complex production runs in the future, using HPC systems connected in grid infrastructures.”
“This is a very good example of how powerful supercomputers at JSC equipped with world-class grid technologies like UNICORE can generate synergies to enable new fields of research. I am proud that JSC is a member of the international UNICORE open source community and leads its development,” explained Prof. Dr. Dr. Thomas Lippert, director of JSC.
The team at SCAI, led by Dr. Marc Zimmermann for the image analysis annotators and by Dr. Juliane Fluck and Dr. Christoph Friedrich for the text analytics part, is currently working on the in-depth analysis of the metainformation generated in the course of this large-scale in silico-experiment.
Related Links:
Fraunhofer Institute for Algorithms and Scientific Computing
Jülich Supercomputing Centre
