Section: Research Program

Computer science – symbolic artificial intelligence

We develop methods that use an explicit representation of the relationships between heterogeneous data and knowledge in order to construct a space of hypotheses. Therefore, our objectives in computer science is mainly to develop accurate representations (oriented graphs, Boolean networks, automata, or expressive grammars) to iteratively capture the complexity of a biological system.

Integrating data with querying languages: Semantic web for life sciences The first level of complexity in the data integration process consists in confronting heterogeneous datasets. Both the size and the heretogeneity of life science data make their integration and analysis by domain experts impractical and prone to the streetlight effect (they will pick up the models that best match what they know or what they would like to discover). Our first objective involves the formalization and management of knowledge, that is, the explicitation of relations occurring in structured data. In this setting, our main goal is to facilitate and optimize the integration of Semantic Web resources with local users data by relying on the implicit data scheme contained in biological data and Semantic Web resources.

Reasoning over structured data with constraint-based logical paradigms Another level of complexity in life science integration is that very few paradigms exist to model the behavior of a complex biological system. This leads biologists to perform and formulate hypotheses in order to interpret their data. Our strategy is to interpret such hypotheses as combinatorial optimization problems allowing to reduce the family of models compatible with data. To that goal, we collaborate with Potsdam University in order to use and challenge the most recent developments of Answer Set Programming (ASP) [58], a logical paradigm for solving constraint satisfiability and combinatorial optimization issues. Our goal is therefore to provide scalable and expressive formal models of queries on biological networks with the focus of integrating dynamical information as explicit logical constraints in the modeling process.

Characterizing biological sequences with formal syntactic models Our last goal is to identify and characterize the function of expressed genes in non-model species, such as enzymes and isoforms functions in biological networks or specific functional features of metagenomic samples. These are insufficiently precise because of the divergence of biological sequences, the complexity of molecular structures and biological processes, and the weak signals characterizing these elements. Our goal is therefore to develop accurate formal syntactic models (automata, grammars, abstract gene models) enabling us to represent sequence conservation, sets of short and degenerated patterns and crossing or distant dependencies. This requires both to determine classes of formal syntactic models allowing to handle biological complexity, and to automatically characterize the functional potential embodied in biological sequences with these models.