Section: Research Program

Implementing methods in software and platforms

Seven platforms have been developped in the team for the last five years: Askomics, AuReMe, FinGoc, Caspo, Cadbiom, Logol, Protomata. Indeed, one of the team's goals is to facilitate interplays between the tools for biological data analysis and integration. Improvements and novelties of these platforms are described in the "software" section. Our platforms aim at guiding the user to progressively reduce the space of models (families of sequences of genes or proteins, families of keys actors involved in a system response, dynamical models) which are compatible with both knowledge and experimental observations.

Most of our platforms are developed with the support of the GenOuest resource and data center hosted in the IRISA laboratory, including their computer facilities [more info]. It worths considering them into larger dedicated environments to benefit from the expertise of other research groups. The BioShadock repository of the GenOuest platform allows one to share the different docker containers that we are developing [website]. The GenOuest galaxy portal of the GenOuest platform now provides access to most tools for integrative biology and sequence annotation (access on demand).

AskOmics platform

Goal Integration and interrogation software for linked biological data based on semantic web technologies [url].

DescriptionAskOmics aims at bridging the gap between end user data and the Linked (Open) Data cloud. It allows heterogeneous bioinformatics data (formatted as tabular files or directly in RDF) to be loaded into a Triple Store system using a user-friendly web interface. AskOmics also provides an intuitive graph-based user interface supporting the creation of complex queries that currently require hours of manual searches across tens of spreadsheet files. The elements of interest selected in the graph are then automatically converted into a SPARQL query that is executed on the users's data.

Originality Our experience is that end users (i) do not benefit for all the information available in the LOD cloud repositories by lack of SPARQL expertise (understandably: they are biologists and most of them do not have an interest in either learning SPARQL nor in learning how to integrate data); (ii) do not contribute their data back to the LOD cloud. Again, they do not have the expertise nor the resources to produce and maintain datasets and the associated metadata as linked data, nor to maintain the underlying server infrastructure. Therefore there is a need for helping end users to (1) take advantage of the information readily available in the LOD cloud for analyzing there own data and (2) contribute back to the linked data by representing their data and the associated metadata in the proper format as well as by linking them to other resources. In this context, the main originality is the graphical interface that allows any SPARQL query to be built transparently and iteratively by a non-expert user.

Application This software was developed in the context of the MirnAdapt (pea-aphid) project in 2016. The tool has been presented to the agriculture communities in conferences [53], [84] and to the Galaxy community [29]. Up to now, more than 10 biological partners team are actually testing and using the prototype software (colza, pea-aphids, copper microbiology, marine biology), and Sanofi has shown its interest to co-develop the tool. Even if its current user base belongs to the bioinformatics community, the scope of AskOmics is domain-independent and has the potential to reach a wider audience related to the Semantic Web community.

AuReMe workspace

Goal Tracable reconstruction of metabolic networks [url].

Description The toolbox AuReMe allows for the Automatic Reconstruction of Metabolic networks based on the combination of multiple heterogeneous data and knowledge sources [64]. It is available as a Docker image. Five modules are composing AuReMe: 1) The Model-management PADmet module allows manipulating and traceing all metabolic data via a local database. [package] 2) The meneco python package allows the gaps of a metabolic network to be filled by using a topological approach that implements a logical programming approach to solve a combinatorial problem [107], [65] and [21] [python package] 3) The shogen python package allows genome and metabolic network to be aligned in order to identify genome units which contain a large density of genes coding for enzymes; it also implements a logical programming approach [60] [python package]. 4) The manual curation assistance PADmet module allows the reported metabolic networks and their metadata to be curated. 5) The Wiki-export PADmet module enables the export of the metabolic network and its functional genomic unit as a local wiki platform allowing a user-friendly investigation [package].

Originality The main added-values are the inclusion of graph-based gap-filling tools that are particularly relevant for the study of non-classical organisms, the possibility to trace the reconstruction and curation procedures, and the representation and exploration of reconstructed metabolic networks with wikis.

Application The tools included in AuReMe have been used for reconstructing metabolic networks of micro and macro-algae [97], extremophile bacteria [16] and communities of organisms [61] in the context of the Idealg, Ciric-omics and IPL Algae-In-Silico projects.

FinGoc-tools

Goal Filtering interaction networks with graph-based optimization criteria.

Description The goal is to offer a set of tools for the reconstruction of networks from genome, literature and large-scale observation data (expression data, metabolomics...) in order to elucidate the main regulators of an observed phenotype. Most of the optimization issues are addressed with Answer Set Programming. 1) The lombarde package enables the filtering of transcription-factor/binding-site regulatory networks with mutual information reported by the response to environmental perturbations. The high level of false-positive interactions is filters according to graph-based criteria. Knowledge about regulatory modules such as operons or the output of the shogen package can be taken into account [39], [38] [web server]. 2) The KeyRegulatorFinder package allows searching key regulators of lists of molecules (like metabolites, enzymes or genes) by taking advantage of knowledge databases in cell metabolism and signaling. The complete information is transcribed into a large-scale interaction graph which is filtered to report the most significant upstream regulators of the considered list of molecules [59] [package]. 3) The powerGrasp python package provides an implementation of graph compression methods oriented toward visualization, and based on power graph analysis. [package]. 4) The iggy package enables the repairing of an interaction graph with respect to expression data. It proposes a range of different operations for altering experimental data and/or a biological network in order to re-establish their mutual consistency, an indispensable prerequisite for automated prediction. For accomplishing repair and prediction, we take advantage of the distinguished modeling and reasoning capacities of Answer Set Programming. [6] [114] [Python package]

Originality The main added-value of these tools is to make explicit the criteria used to highlight the role of the main regulators: the underlying methods encode explicit graph-based criteria instead of relying on statistical approaches. This makes it possible to explain local relationships and patterns within interaction graphs by explicit biological relationships.

Application The tools have been used to figure out the main gene-regulators of the response of porks to several diets in [74], [76] and [18]. The tools were also used to to decipher regulators of reproduction for the pea aphid, an insect that is a pest on plants [85], [119].

Caspo software

Participant : Anne Siegel.

Goal Studying synchronous boolean networks [url]

Description Cell ASP Optimizer (Caspo) constitutes a pipeline for automated reasoning on logical signaling networks. The main underlying issue is that inherent experimental noise is considered, many different logical networks can be compatible with a set of experimental observations (see [106] and [22]). It is available as a Docker container. Five modules are composing Caspo: 1) the Caspo-learn module performs an automated inference of logical networks from experimental data allows for identifying admissible large-scale logic models saving a lot of efforts and without any a priori bias [115] and [78]. 2) The Caspo-classify, predict and visualize modules allows for classifying a family of boolean networks with respect to their input-output predictions [78]. 3) The Caspo-design module designs experimental perturbations which would allow for an optimal discrimination of rival models in a family of boolean networks [116]. 4) The Caspo-control module identifies key-players of a family of networks: it computes robust intervention strategies that force a set of target species or compounds into a desired steady state [80]. 5) The Caspo-timeseries module to take into account time-series observation datasets in the learning procedure [94] [python package and docker container].

Originality The Caspo modules provide friendly and efficient solutions to problems that were previously addressed in theoretical papers with MILP programs. The main advantage is that is enables a complete study of logical network without requiring any linear constraint programs.

Application The Caspo tool was initiated in the framework of the BioTempo project. Caspo-learn has been included as a module to learn logical networks from early steady-state data in CellNopt, a generic platform which implements several methods for learning and studying signaling networks are different modeling levels (from logical models to numerical models).

Cadbiom package

Goal Building and analyzing the asynchronous dynamics of enriched logical networks [url]

Description Based on Guarded transition semantic, the Cadbiom software provides a formal framework to help the modeling of biological systems such as cell signaling network. It allows synchronization events to be investigated in biological networks [40]. It is available as a Docker image. Three modules are composing Cadbiom: 1) The Cadbiom graphical interface is useful to build and study moderate size models. It provides exploration, simulation and checking. For large-scale models, Cadbiom also allows to focus on specific nodes of interest. 2) The Cadbiom API allows a model to be loaded, performing static analysis and checking temporal properties on a finite horizon in the future or in the past. 3) Exploring large-scale knowledge repositories, the translations of the large-scale PID repository (about 10,000 curated interactions) have been translated into the Cadbiom formalism.

Originality Model-checking approaches applied to Boolean networks [81] or multivalued networks [91] allow the trajectories of the system to be entirely studied but they can only be applied to small-size networks. On the contrary, Cadbiom is able to handle large-scale knowledge databases.

Application The Cadbiom tool was applied to study the regulators of the TGF-$\beta$, a gene that controls liver fibrosis [40] in the framework of the TGFSysBio project. The study of its predictions also enabled large-scale knowledge databases (PID) to be curated [25].

Logol software

Goal Complex pattern modelling and matching [url]

Description The Logol toolbox is a swiss-army-knife for pattern matching on DNA/RNA/Protein sequences, using a high-level grammatical formalism to permit a large expressivity for patterns [50]. A Logol pattern can consist in a complex combination of motifs (such as degenerated strings) and structures (such as imperfect stem-loop ou repeats). Logol key features are the possibilities to divide a pattern description into several sub-patterns, to model long range dependencies, to enable the use of ambiguous models or to permit the inclusion of negative conditions in a pattern definition. The LogolMatch parser takes as input a biological sequence and a grammar file. It returns a XML file containing all the occurrences of the pattern in the sequence with their parsing details. The input sequences can be genomes from biological banks.

Originality Many pattern matching tools exist to efficiently model specific types of patterns: vmatch [82], patmatch [121], cutadapt [89], scoring matrix or profile HMMs [44], [71]. The main advantage of Logol is its very large expressivity. It encompasses most of the features of these specialized tools and enables interplays between several classes of patterns (motifs and structures).

Application The Logol tool was applied to the detection of mutated primers in a metabarcoding study [41], [42] or to stem-loop identification (e.g. in CRISPR (http://crispi.genouest.org/, https://hal.inria.fr/hal-00643408) [103], [50]). Ongoing application is the search for transposable elements in the human genome in the context of a colorectal cancer study (Preprint: http://www.biorxiv.org/content/early/2017/03/09/115030). Logol strongly supported the study of the LXR-$\alpha$ targets in the framework of the FatInteger project.

Protomata-suite

Goal Expressive pattern discovery on protein sequences [url]

DescriptionProtomata is a machine learning suite for the inference of automata characterizing (functional) families of proteins from available sequences. Based on partial and local alignments, Protomata learns precise characterizations of the families of proteins, allowing new family members to be predicted with a high specificity. Three main modules are integrated in the Protomata-learner workflow are available as well as stand-alone programs: 1) paloma builds partial local multiple alignments, 2) protobuild infers automata from these alignements and 3) protomatch and protoalign scans, parses and aligns new sequences with learnt automata. The suite is completed by tools to handle or visualize data and can be used online by the biologists via a web interface on Genouest Platform. It is actively maintained (version v2.1 was released in April 2017) and we are scheduling a new major version with enhanced scoring schemes that we have proposed [105].

Originality The main specificity is that the power of characterization is beyond the scope of classical sequence patterns such as PSSM (e.g. MEME suite [43]), Profile HMM (e.g. HMMER package [71]), or Prosite Patterns [111] allowing new family members to be predicted with a high specificity.

Application The Protomata tool is used both to update automatically the Cyanolase database [62] and, when combined to Formal Concept Analysis, to automated enzyme classification, such as the HAD superfamily of proteins [67] in the framework of the Idealg project.