PractiKPharma
Practice-based evidences for actioning Knowledge in Pharmacogenomics
ANR-15-CE23-0028 -- http://practikpharma.loria.fr
2016-2021
OBJECTIVES
PractiKPharma is a collaborative research project funded by the French National Research Agency (ANR). It studies computer sciences approaches to extract, compare, validate state of the art knowledge in the biomedical domain of Pharmacogneomics.
Pharmacogenomics studies how genetics impacts drug response phentoypes.
Units of knowledge in PGx typically have the form of ternary relationships gene variant–drug–adverse event, and can be formalized in various manner, in particular with biomedical ontologies.
1. Knowledge Extraction from reference databases & the biomedical literature
We extract from open-access biomedical articles (PubMed citations), strongly typed PGx relationships between fine grained PGx entities, i.e., gene variant, drug, adverse event.
We focus on the description of phenotypes related to drug responses in term of signs and symptoms e.g., “anemias”, or “low RBC”.
A challenge for the extraction of PGx knowledge is to associate with PGx relationships a validation level on how hypothetical a relationship is. For evaluating knowledge extraction quality we use the content of the Pharmacogenomic Knowledge Base (PharmGKB) developed at Stanford.
2. Knowledge Extraction from Electronic Health Records (EHRs)
We treat Electronic Health Record data from the HEGP and annotate their content with biomedical ontologies. Because EHRs are in French, we reuse and adapt the SIFR Annotator to annotate French clinical narratives with French ontologies.
3. Comparison of state-of-the-art and observational knowledge
Knowledge previously extracted from literature and clinical notes are from different nature and come from multilingual data. We develop a framework for comparing those knowledge units and detect concurring, ambiguous and conflicting knowledge units. We study Formal Concept Analysis to establish correspondences between knowledge representations and data sources. We consequently map ontologies natively used in both worlds (literature and EHRs), considering that they can be either in English or in French.
4. Explaining PGx knowledge using omics data
We consider several omics databases (e.g., DrugBank, Uniprot, KEGG) to investigate molecular mechanisms that may explain adverse reactions of pharmacogenomic drugs. Some of these databases are available as Linked Open Data, facilitating their reuse, some are not. Once these databases connected, we will evaluate novel data analysis methods to provide elements of explanation of how PGx genes impact drug response.
RESULTS
Publications
10 journals, 13 international workshops or conferences, 4 national workshops or conferences (as of May 2019)
We published in international journals (Journal of Biomedical Semantics, Gigascience, Scientific Reports, BMC Bioinformatics, Scientific Data), in international conferences (EKAW, ISMIS, MedInfo), in international peer-reviewed workshops (LOUHI, NETTAB, BioOntologies, etc.) and a popular article in ERCIM News magazine.
Nouvelle
Pharmacogenomic Linked Open Data (PGxLOD)
Linking PGx data
We started with six data sources which focuses on drugs (DrugBank), genes (ClinVar, DisGeNET) and drug responses (Side, MediSpan) and added pharmacogenomic knowledge units extracted from PharmGKB and the litterature. These data sources were mapped and transformed into a single RDF graph available at https://pgxlod.loria.fr. PGxLOD is intended to host pharmacogenomic-related data and knowledge units of various provenance. PGxLOD paper.
Pharmacogenomic Ontology (PGxO)
A simplistic representation of PGx relationships and their provenance
To structure and allow the coexistence of pharmacogenomic knowledge elements of diverse origins within PGxLOD, we propose a minimal ontology, called PGxO. This simple schema enables both representing pharmacogneomic knowledge units and documenting their provenance. PGxO paper.
Pharmacogenomic Corpus (PGxCorpus)
Deep Learning for knowledge extraction
We aim at enabling deep learning to extract PGx relationships from text. Because these approaches require large sets of training data we have built a corpus, called PGxCorpus, from PubMed articles currently made of 945 sentences with >80% with a PGx relationship (2875) manually annotated by 11 annotators.
Meanwhile, we are investigating transfer learning approaches to enable PGx relationship extraction.PGxCorpus paper.
SIFR Annotator
Semantic annotation of clinical notes
We have adapted the French Annotator developed within the SIFR project to annotate French clinical text. We added enhanced functionalities such as scoring, detection of context (negation, experiencer, temporality), new output formats, and coarse-grained concept recognition.
Clinical Data Bioinformatics Workflow
Facilitate clinical records process
We developed a bioinformatics pipeline using NextFlow to facilitate the processing and populating of HEGP Electronic Health Records. The resulting workflow enables the design of reproducible studies associated with metadata and parametrization. Molecular biologists use it to annotate tumor variants.
Software Developments
Our software is open source and available on GitHub: https://github.com/practikpharma
PractiKPharma scientific meetings
PractiKPharma plenary meetings
Invited: Clement Goerhs (ERIAS/C-NAPPS)
Other co-organized events
External collaborations
Tutorials & Presentations
Publications & Software
Publications related to the project are available on HAL,
Code is on GitHub.
Hackathon
PractiKPharma has participated to the BioHackathon 2018 in Paris. Our team worked on exploring pharmacogenomic linked open data.
Prize of the EGC 2022 PhD Thesis
The EGC association granted his 2022 PhD prize to Pierre Monnin for his thesis untitled "Matching and mining in knowledge graphs of the Web of data : Applications in pharmacogenomics" prepared within PractiKPharma. Congrats Pierre!
TEAM
Pierre Monnin
PhD student
LORIA
Anita Burgun
Professor
HEGP
JOBS
Research engineer position (closed)
Web architect – Semantic web technologies – Text and data mining, LIRMM, Montpellier
Postdoctoral position (closed)
Postdoctoral position (closed)
PhD position (closed)
CONTACT
SUPPORT
OTHER SUPPORT