Pharmacoracle

Pharmacoracle - a REST API based on the OWLAPI

The Pharmacoracle contains OWL ontologies that are based on CPIC recommendations provided by the PharmGKB API, and which have been further curated to satisfy the needs of the Department of Pharmacology at Oslo University Hospital.

The Pharmacoracle was implemented as an attempt to understand the field of pharmacogenomics, based on data from PharmGKB, and solving the same problem as PharmGKBs PharmCAT software, but in the context of the technology and clinical recommendations of our clinic. It may be sensible to integrate PharmCAT in the Pharmacoracle, as soon as PharmCAT becomes the gold standard for VCF to PGx recommendations.

Queries to Pharmacoracle are written in the Manchester syntax in the same way as for the DL Query Tab of Stanford's desktop Protégé. An advantage of this approach is that it replaces the need of a separate rules engine (rules are to some extent built into the OWL semantics). A disadvantage is that the performance in a production environment may be suboptimal (slow). I believe that the standard way to use OWL is to make a triplet store database (RDF database, e.g. JENA) based on the OWL ontology. The rules can then be written as SPARQL queries to this database. It is possible that there are more lessons to be learned from the PGx clinical decision support ontology of Mathias Samwald, which is at the core of the Ontology-Based, Mobile-Optimized System for Pharmacogenomic Decision Support at the Point-of-Care of the UPGx project. They use the TrOWL reasoner instead of the HermIT reasoner that is at the core of Pharmacoracle.

The Pharmacoracle infrastructure is written in Python. But the OWLAPI is mainly JAVA technology, and making the Pharmacoracle totally JAVA, might speed up things. A link between OWL and Python is provided by Jean Baptiste Lamy's Owlready2, which we use to translate PharmGKB's JSON-LD to OWL, but for the Pharmacoracle the OWLAPI is run separately.

The Pharmacoracle is used in our PGx analysis script Pharmacolyzer, but can also be used directly.

Hints on how to use it (see also example section below)

In order to write queries to an OWL ontology we need to know the names of the OWL elements, notably OWL Properties (e.g. has_haplotype) and and OWL Classes (e.g. TPMTstar3A). (The Oslo University Hospital PGx OWL ontology that is used by the Pharmacoracle is still work in progress, but it is possible to have a closer, off-line look at the latest OWL version of the GRCh37 star allele definitions, which is best studied in Stanford's Desktop Protégé). In addition we need to know the Manchester expressions of cardinality that are used to connect OWL Properties and OWL Classes, in particular:

Cardinality word	Cardinality	Explanation
some		Connects a property to at least one class, e.g. has_haplotype some TPMTstar3A
exactly	number	Connects a property to an exact number of classes, e.g. has_haplotype exactly 1 TPMTstar3A
min	number	Connects a property to a minimum number of classes, e.g. has_haplotype min 1 TPMTstar3A

For each query we add a code word to indicate the relationship between our query and the desired answer (cf. the DL Query tab in the desktop version of Stanford's Protégé), as explained in this table:

Code word	Relationship
SubClasses	Get elements that contain the query (but not equivalent to the query)
DirectSubClasses	Get only the SubClasses that are closest to the query
EquivalentClasses	Get only elements equivalent to the query
SuperClasses	Get elements that contain parts of the query (but not the entire query and not equivalent)
DirectSuperClasses	Get only the SuperClasses that are closest to the query

In order to query the Pharmacoracle automatically, we have made a Python script Pharmacolyzer that returns dosing recommendation based on a pharmacotyped patient VCF.

Examples of use

Find the PGx alleles that contain the variant snpC>T at chr6:18139228

curl https://www.pgx.no/api/pharmacoracle -d "query=SubClasses:snpCtoT_at some chr6_18139228"

Find the PGx allele that is identical to the variants snpC>T at chr6:18139228 and snpT>C at chr6:18130918

curl https://www.pgx.no/api/pharmacoracle -d "query=EquivalentClasses:(snpCtoT_at some chr6_18139228) and (snpTtoC_at some chr6_18130918)"

Find the PGx alleles that contain either snpC>T at chr6:18139228 or snpT>C at chr6:18130918

curl https://www.pgx.no/api/pharmacoracle -d "query=DirectSuperClasses:(snpCtoT_at some chr6_18139228) and (snpTtoC_at some chr6_18130918)"

Find the Azathioprine TPMT functional status of the TPMT*3B allele:

curl https://www.pgx.no/api/pharmacoracle -d "drugs=azathioprine&query=SuperClasses:has_haplotype some TPMTstar3B"

Find the Azathioprine dosing recommendations for a patient with two Azathioprine TPMT No Function alleles:

curl https://www.pgx.no/api/pharmacoracle -d "drugs=azathioprine&query=EquivalentClasses:has_phenotype some (has_function exactly 2 AzathioprineTPMT_NoFunction)"