Integrate and query local datasets and distant RDF data with AskOmics using Semantic Web technologies

Contributors

Authors: Xavier Garnier Anthony Bretaudeau Anne Siegel Olivier Dameron Mateo Boudet

Questions

• What is the Semantic Web and how can it help integrating and querying data?

• How can AskOmics help benefiting from Semantic Web technologies without having to write RDF or SPARQL code?

Objectives

• Understand the basics of RDF and SPARQL

• Learn how input data must be structured to be integrated with AskOmics

• Learn how to connect distant SPARQL endpoints to local data with AskOmics

How to explore data

Requirements

Study of biological mechanisms requires to:

• integrate multiple data sources (differential expression results, genome annotation, remote protein database)
• query them to answer a biological question (which genes are over-expressed in a condition, and what are the proteins coded by these genes)

.image-00[ ]

What is the Semantic Web?

Semantic Web

Set of recommendations to integrate data, to integrate domain knowledge and to perform query and reasoning.

• Resource Description Framework (RDF): annotate data
• RDFS + OWL: represent knowledge (data description)
• SPARQL Protocol and RDF Query Language (SPARQL): query data

RDF

• RDF is for describing resources (the R in RDF)
  • resources are identified by URIs (nextprot:P01137, taxon:9606)
• describing (D in RDF) a resource is representing it explicitly
  • its attributes (nextprot:P01137 :hasSequence "MPPSGLRLLL...")
  • its relations to other entities (nextprot:P01137 :hasTaxon taxon:9606)
  • its descriptions (aka classes) (nextprot:P01137 :is nextprot:Protein)

RDF: Set of triples

• RDF is represented by triples (subject, predicate, object)
  • Subject: the resource being described
  • Predicate: the relation (from subject to object)
  • Object: a value of the predicate for the subject

.image-01[ ]

nextprot:P01137 :hasTaxon taxon:9606 .
nextprot:P01137 :hasSequence "MPPSGLRLLL" .

RDF: triples form a labeled directed graph

# Description
nextprot:P01137 rdf:type nextprot:Protein .
taxon:9606 rdf:type nextprot:Organism .
# Data
nextprot:P01137 :hasTaxon taxon:9606 .
nextprot:P01137 :hasSequence "MPPSGLRLLL" .

.image-02[ ]

SPARQL

• The SPARQL language is a set of triple patterns with variables (?variable_name)

SELECT ?gene
WHERE {
    ?gene rdf:type :Gene .
    ?gene :hasTaxon taxon:9606 .
}

• All ?gene with rdf:type :Gene and with :hasTaxon taxon:9606
• In other words, the query returns all the human genes

SPARQL: entity matching allow federated queries

.pull-left[

• Using the same identifier for the same entity (entity matching) across multiple datasets allows federated queries

• Federated SPARQL queries provide unified querying capabilities over multiple datasets as if they were a single virtual graph

]

.pull-right[ .image-03[ ] ]

What is AskOmics?

AskOmics

Web software for data integration and query using Semantic Web. The main functionalities are:

• Convert of multiple data formats into RDF triples and store them in a local triplestore
• Generate complex SPARQL queries using a user-friendly interface
• Support external SPARQL endpoint to cross-reference integrated data with remote data

AskOmics can be used as a standalone software, or with Galaxy

Data integration with AskOmics

Local data integration

• RDF and SPARQL are good infrastructures to describe and query biological datasets, but most of them are still stored on flat files like CSV/TSV and GFF.

• AskOmics converts multiple structured data formats into RDF triples

  • CSV/TSV
  • GFF
  • BED

AskOmics generates the graph of data and the abstraction

AskOmics uses the file structure (e.g. header of TSV files) to generate the graph of data description: the abstraction

.image-04[ ]

The rest of the files is converted to RDF triples that correspond to the data.

Distant RDF data integration

• Some public databases (e.g. neXtProt) provide RDF data through a SPARQL endpoint (public access for RDF data)

• To connect with a remote SPARQL endpoint, AskOmics needs its RDF abstraction
• The abstraction can be generated with abstractor

pip3 install abstractor
abstractor -s https://sparql.nextprot.org/sparql -o nextprot_abstraction.ttl -f turtle

• This abstraction can then be uploaded into AskOmics as a standard file

Query multiple data sources with AskOmics

Query composition

• Users navigate through the abstraction of local and remote data and create a path that represents a query

.image-06[ ]

• The query is converted to SPARQL code that is executed on the local and remote RDF data
• Results are returned and can be downloaded

Key Points

• RDF and SPARQL are Semantic Web technologies that come useful for data integration and querying but the technical aspects may deter end-users
• AskOmics is a web platform for data integration and query using Semantic Web in a user-friendly way

Thank you!