Research and development of innovative in-vitro diagnostic (IVD) solutions in infectious diseases require to federate up-to-date knowledge from several fields known for their complexity and their constant evolution: medical practices, microbiology and system and software engineering [6, 11, 12]. To tackle the inherent complexity of such multidisciplinary R&D projects, modern information technologies now offer powerful environments which can be leveraged to facilitate information sharing between corporate experts. This is key to ensure semantic alignments, information retrieval and then to foster decision making within the projects. The advent of almost mature semantic technologies together with international standards bring the possibility to create enterprise compliant knowledge bases. The major challenge is then to gather and link all the information from distinct and heterogeneous sources in a frequently updated and fully searchable resource. Ideally, for IVD projects, such a resource would allow for instance to map unmet needs onto current medical practices in infectious diseases, to facilitate comparison of results from different technologies, or to gather and maintain pathogen-related knowledge. Towards this goal, we benefited from the recent efforts from the biomedical and bioinformatics communities which have been early adopters of the promising web 3.0 functionalities; multiple public data resources have developed and released domain specific ontology models or SPARQL endpoints [5, 8, 14]. Taking advantage of these semantic components we deployed on the company intranet BioPedia, a private collaborative semantic web platform carrying a cross domain knowledge base dedicated to human pathogens. The knowledge is stored on a triplestore while a wiki-based interface allows to create powerful faceted queries.
The current architecture of Biopedia is based on a central triplestore interfaced
with sparql 1.1 compliant 4store [
The triplestore is populated with mashed up data mapped on the ontology. The
mashup of data from heterogeneous sources includes ontology alignments, terms
mapping or bridges between synonym concepts from the company and from
public sources. BioTaxon domain contains bridges translating corporate identifiers
to NCBI [
A mapping between Strain and Genome was built in order to federate public
genome data from PATRIC and from our internal genome database and thus
populate the BioSeq domain. Genome sequences can be processed to provide
annotation that can be used as one source to populate the Locus class. Here,
Loci that are registered as Resistance determinants can give a very valuable
information about the strain ability to resist to antimicrobials. BioGraM is the
alignment result between our corporate master data knowledge base and the
Comprehensive Antibiotic Resistance Database (CARD) [
The current triplestore contains more than 14 million triples linking the four domains of BioPedia and allowing to infer new knowledge (Figure 2). The bridge between the taxonomy nomenclatures of NCBI and our corporate reference taxonomy was built using the taxon translator tools we developed. However 10% of the corporate labels could not be mapped on NCBI taxon labels (Figure 2 A. 3)). This is partially due to a shift of nomenclature versions between sources. Indeed, because labels are not standardized, NCBI can still uses Fluoribacter bozemanae when ICSP suggests Legionella bozemanae. For these labels, the bridge is manually completed by our expert taxonomy curator.
As shown on Figure 2 A. 1), public strains were much more reduced when gathered into clusters as there are very connected data while private strain have fewer strain cross-references to clusterize. Among the 114,383 strain clusters in which at least one strain belongs to our corporate collection, 8% were allowed to link much more metadata such as public genomes. The validation process also highlighted 2% of clusters whose strains did not share a common taxon identifier. A sparse matrix is then used to help the curator to identify the incriminated vertices.
The integration of this content in the semantic web portal BioPedia provides powerful querying tools such as a hierarchical browser to navigate within the taxonomy classification or faceted searches based on semantic properties. Together with the sparql facilitator provided by the LinkedWiki extension, this allows us bringing a solution for R&D project teams to (i) easily federate all the available data generated so far for any pathogen stored into the global strain collection or (ii) create reference strain panels based on various criteria depending on targeted diagnostic applications (Figure 2 B.). The microbial knowledge base provides global and uniform knowledge of the company strain collection and links it to many infectious diseases oriented public metadata, such as resistance to antimicrobials or genomes. This work gives an enriched overview of this strain collection and connects it to the achievements of the scientific community.
Then, as a side-benefit, linking data from several sources through a semantic store is of great help to improve data quality. Indeed in the mashups, sibling concepts from heterogeneous information streams are blended together and this new closeness drastically highlights the discrepancies. The data curation, even semi-automatic, is time-consuming but mandatory to build a trustworthy reference knowledge base on which powerful queries can be launched and reference datasets can be exported with confidence. The resulting collaborative semantic web service makes possible to connect heterogeneous data in a corporate way. As the access to data is centralized, it avoids data silo and data tomb often caught out in excel spread-sheets without associated metadata. Moreover the collaborative aspect of this system encourages scientific experts to complete missing information that are then validated by a moderator, thus participating to the enrichment and quality increase of the knowledge base.