Jan Grzegorzewski
Humboldt-University Berlin, Institute for Theoretical Biology, Berlin
Objectives:
A multitude of pharmacokinetics studies has been published. However, due to the lack of an open database, pharmacokinetics data, as well as the corresponding meta-information, have been difficult to access. We present PK-DB (https://pk-db.com), an open database for pharmacokinetics information from clinical trials including pre-clinical research. PK-DB provides curated information on (i) characteristics of studied patient cohorts and subjects (e.g. age, body weight, smoking status); (ii) applied interventions (e.g. dosing, substance, route of application); (iii) measured pharmacokinetic time-courses; (iv) pharmacokinetic parameters (e.g. clearance, half-life, area under the curve). Key features are the representation of experimental errors, the normalization of measurement units, annotation of information to biological ontologies, calculation of pharmacokinetic parameters from concentration-time profiles, a workflow for collaborative data curation, strong validation rules on the data, computational access via a REST API as well as human access via a web interface. PK-DB enables meta-analysis based on data from multiple studies and data integration with computational models. A special focus lies on meta-data relevant for individualized and stratified computational modeling with methods like physiologically based pharmacokinetic (PBPK), pharmacokinetic/pharmacodynamic (PK/DB), or population pharmacokinetic (pop PK) modeling.
Methods:
The PK-DB backend is implemented in Python using the Django framework with Postgres as the underlying database system. For fast, full-text search most data is indexed with Elasticsearch. The provided REST API uses the Django-rest-framework with endpoints accessible from https://pk-db.com/api/. The web frontend (https://pk-db.com) is implemented in JavaScript based on the Vue.js framework interacting with the backend via the REST API. The complete PK-DB stack is distributed as docker-containers. PK-DB is licensed under GNU Lesser General Public License version 3 (LGPL-3.0) with source code available from https://github.com/matthiaskoenig/pkdb.
Results:
PK-DB provides curated information on (i) characteristics of studied patient cohorts and subjects (e.g. age, body weight, smoking status); (ii) applied interventions (e.g. dosing, substance, route of application); (iii) concentration-time curves; and (iv) parameters measured in PK studies (e.g. clearance, half-life, and area under the curve). The focus so far has been on substances applied in dynamical liver function tests and studies of glucose metabolism. PK-DB consists of 332 studies containing 869 groups, 1808 individuals, 997 interventions, 28220 outputs and 2118 time-courses related to caffeine, glucose, codeine, morphine, atorvastatin, midazolam or paracetamol (Feb 2020).
Accessibility of data for computational modeling and data science
All data is available via REST endpoints allowing for simple integration of PK-DB data into existing workflows, e.g., for the building of computational models. The major advantage of a REST API as a central access point to the database is that it can be accessed from various clients independent of the programming language. The use of PK-DB data is facilitated by annotation of biological and medical concepts to respective ontologies. This enables the integration with additional data sets or computational models based on the semantic information, e.g., substances are annotated to ChEBI, and diseases to ncit, hp, doid, and mondo. A special focus lies on meta-data for individualized and stratified computational modeling with methods like physiologically based pharmacokinetic (PBPK), pharmacokinetic/pharmacodynamic (PK/DB), or population pharmacokinetic (pop PK) modeling.
Extensibility and generalizability
The PK-DB data model is not limited to a specific problem domain but allows simple extensions to other fields and experimental data sets, within the overall area of pharmacokinetics. Examples are extensible types for the group or individual characteristics currently represented in the database. Additional types can easily be added to cover the important information for a given problem domain.
Unit and data normalization
A key challenge in using data for computational modeling and data science are non-standardized units coming from different data sets. It requires time-consuming retrieval of this information from the literature and error-prone conversion of units and corresponding data. PK-DB provides a solution to this issue. During upload the data is harmonized, e.g., data is converted between molar units and gram, using thereby the molecular mass of the respective substances based on its ChEBI information.
Representation of time-course data
The main measures in pharmacokinetics studies are concentration-time curves of the administered substance and its metabolites after biotransformations. These time-courses are crucial for kinetic modeling, e.g., using physiologically based pharmacokinetic (PBPK) or pharmacokinetics/pharmacodynamics (PK/PD) models. Consequently, a central focus was on storing and analyzing such data efficiently
Data quality
Strong validation (e.g. of categoricals), minimum relevant information, instance cross-referencing and correct unit-dimensions ensure high quality of the curated data. Non-obvious curation mistakes (or respective errors in the reporting of the data) can be addressed by outlier identification in subsequent meta-analyses.
Conclusions:
We present PK-DB, a database for the representation of pharmacokinetics data from clinical trials and pre-clinical research. Data is either curated from the literature or from available raw data. The main focus of PK-DB is to provide high-quality pharmacokinetics data in combination with required meta-information for computational modeling and data integration, i.e., (i) characteristics of studied patient collectives and individuals; (ii) applied interventions (e.g. dosing, substance, route of application); (iii) measured pharmacokinetics information and time courses.
Important features are the representation of experimental errors and variation, the representation and normalization of units, annotation to biological ontologies, and calculation of pharmacokinetics information like apparent clearance, half-life, or area under the curve (AUC) from timecourse data.
References:
[1] PK-DB: PharmacoKinetics DataBase for Individualized and Stratified Computational Modeling Jan Grzegorzewski, Janosch Brandhorst, Dimitra Eleftheriadou, Kathleen Green, Matthias König bioRxiv 760884; DOI: https://doi.org/10.1101/760884
Reference: PAGE () Abstr 9502 [www.page-meeting.org/?abstract=9502]
Poster: Oral: Methodology - New Tools