Section: New Results

Scalable methods to query data heterogenity

Participants : Emmanuelle Becker, Lucas Bourneuf, Olivier Dameron, Xavier Garnier, Vijay Ingalalli, Marine Louarn, Yann Rivault, Anne Siegel.

Increasing life science resources re-usability using Semantic Web technologies [E. Becker, O. Dameron, X. Garnier, V. Ingallali, M. Louarn, Y. Rivault, A. Siegel ] [25], [18], [29], [31], [23], [27], [28]. Our work was focused on assessing to what extent Semantic Web technologies also facilitate reproducibility and reuse of life sciences studies involving pipelines that compute associations between entities according to intermediary relations and dependencies.

• We followed on 2018 action exploratoire Inria by studying possible optimizations for federated SPARQL queries [31]

• We considered a case-study in systems biology ([Regulatorycircuits link]), which provides tissue-specific regulatory interaction networks to elucidate perturbations across complex diseases. We relied on this structure and used Semantic Web technologies (i) to integrate the Regulatory Circuits data, and (ii) to formalize the analysis pipeline as SPARQL queries. Our result was a 335,429,988 triples dataset on which two SPARQL queries were sufficient to extract each single tissue-specific regulatory network.

• A second case-study concerned public health data for reusing electronic health data, selecting patients, identifying specific events and interpreting results typically requires biomedical knowledge  [64]. We developed the queryMed R package [18], [29]. It aims to facilitate the integration of medical and pharmacological knowledge stored in formats compliant with the Linked Data paradigm (e.g. OWL ontologies and RDF datasets) into the R statistical programming environment. We showed that linking a medical database of 1003 critical limb ischemia (CLI) patients to ontologies allowed us to identify all the drugs prescribed for CLI and also to detect one contraindicated prescription for one patient. We also investigated temporal models of care sequences for the exploration of medico-administrative data as part of Johanne Bakalara's PhD, supervized with Thomas Guyet (Lacodam) and Emmanuel Oger (Repères).

• We pursued the development of AskOmics [27]. Version 3 adds the capability to generate the graph of entity types (aka abstraction) from typed RDF datasets, improved management of entity hierarchies and support for federated queries on external SPARQL endpoints such as UniProt and neXtProt.

Graph compression and analysis [L. Bourneuf] [26], [24]. Because of the increasing size and complexity of available graph structures in experimental sciences like molecular biology, techniques of graph visualization tend to reach their limit.

• We developed the Biseau approach, a programming environment aiming at simplifying the visualization task. Biseau takes advantage of Answer Set Programming and shows as a use-case how Formal Concept Analysis can be efficiently described at the level of its properties, without needing a costly development process. It reproduces the core results of existing tools like LatViz or In-Close.

• We formalized a graph compression search space in order to provide approximate solutions to the NP-complete problem of computing a lossless compression of the graph based on the search of cliques and bicliques. Our conclusion is that the search for graph compression can be usefully associated with the search for patterns in a concept lattice and that, conversely, confusing sets of objects and attributes brings new interesting problems for FCA.