Researchers at Madrid's UPM Facultad de Informatica have developed a new tool that can uncover and automatically classify bioinformatics resources from the scientific literature.
Bioinformatics resources have proliferated in the years since the end of the Human Genome Project, obliging researchers to spend a quite a lot of time browsing the web in searches. The Universidad Politecnica de Madrid' s Biomedical Informatics Group (GIB) based at the Facultad de Informatica has developed an innovative methodology, the first capable of discovering and automatically classifying bioinformatics resources from the scientific literature.
Nowadays the scientific community has access to many online bioinformatics resources. This number grows exponentially day by day.
Led by Professor Victor Maojo, a team of researchers from the GIB (including Guillermo de la Calle, Miguel Garcia-Remesal, Diana de la Iglesia, and Stefano Chiesa) have developed an innovative methodology designed to discover, retrieve, and automatically classify bioinformatic resources from specialized scientific literature. The developed index of resources is freely available via the web application BioInformatics Resource Inventory (BIRI).
The methodology is based on natural language processing and artificial intelligence techniques used to retrieve and automatically classify key information contained in scientific articles — primarily abstracts. Each article is analyzed morphologically, syntactically, and semantically in search of a series of set patterns that are able to automatically identify the names, functionality, access URL, and in some cases, the resource inputs and outputs without user intervention.
Additionally, the resources are classified by two dimensions: the application domain (for example, DNA or proteins) and the category (functionality/type) of the resource (for example, alignment, database or annotation). For the purposes of classification, the application uses a taxonomy of domains and categories specially designed for this purpose and based on other existing taxonomies.
To validate the methodology, the UPM group ran a preliminary experiment on 400 articles indexed in the ISI Web of Knowledge. A search was run with the "bioinformatics resources" string and selected the top 392 most relevant articles by impact factor. The others articles were unrelated to bioinformatics resources and were entered as a control group to verify method robustness. A total of 376 names of resources were automatically retrieved from the above set of resources. This amounts to a success rate of almost 95%.
Additionally, a web-services-based web application has been built for the scientific community to use to access the index and search resources by name, category and domain.
The key advantage of this method over existing resource indexes is that it is automatically created and updated. As it is a general-purpose methodology, it is being applied as part of the European ACTION-Grid project, the first European Grid Computing, Biomedical Informatics and Nanoinformatics Initiative, coordinated by Professor Victor Maojo.