=Paper=
{{Paper
|id=None
|storemode=property
|title=OWL Representation of Drug Activities on Biological Systems
|pdfUrl=https://ceur-ws.org/Vol-897/ecs_2.pdf
|volume=Vol-897
|dblpUrl=https://dblp.org/rec/conf/icbo/Croset12
}}
==OWL Representation of Drug Activities on Biological Systems==
https://ceur-ws.org/Vol-897/ecs_2.pdf
OWL Representation of Drug Activities
on Biological Systems
Author Samuel Croset
Supervisors Dietrich Rebholz-Schuhmann
Studies/Stage PhD student, second year
Affiliation European Bioinformatics Institute
E-Mail croset@ebi.ac.uk
Aims and Objectives of the Research
The pharmaceutical industry is currently facing a decrease in productivity. Despite the
increasing investments made in research and development, the number of newly approved
drugs on the market is not growing. To overcome this issue, a strategy, drug-repurposing,
aims at re-assigning already approved drugs towards new disease indications. This
approach has witnessed numerous successes such as the safe rehabilitation of withdrawn
drugs such as Duloxetine for a totally different use than the original one [1]. Repurposing
opportunities can arise from fortunate situations or advantageous side-effects, but
researchers are developing numerous methodologies to identify and predict potentially
relevant cases in a systematic way. We are working on such a methodology, based on
description logic. Our method aims at integrating knowledge from the molecular to the
phenotypic level, with data coming from biological pathways repositories and Gene Ontology
(GO) annotations. The method would offer a formal framework on which drug re-purposing
hypotheses could be tested and explained via a reasoner running on the top of the integrated
knowledge.
Justification for the Research Topic
Motivations behind the Description Logic approach: The vast majority of drug re-profiling
approaches are based on previous knowledge stored in public databases. From this
information, statistical methods can process the data and identify the significant re-purposing
opportunities based on features association. It is however difficult to provide a pragmatic
explanation behind the predictions made in order to support further improvements for the
development of the drug. Moreover, the outcomes strongly depend on the choice of the
features used [2]. A different approach, developed within the systems biology field, models
the dynamics of a set of molecules known to be involved in a disease, in order to predict the
action of a drug. Such techniques provide solid arguments to explain the mechanism of
action of the compound, but are sensitive to missing knowledge or parameters. Moreover it is
difficult to integrate additional data from the phenotypic level, as the modelling is often built
upon chemical reactions only [2]. Our approach is complementary yet different to the ones
previously described. It uses description logic to capture the rational thinking behind the
design of a drug towards a disease. From a set of biological facts and based on a series of
deductions, our framework could explain the logical reasons why a drug would influence a
phenotype. Facts can be extracted from ontologies and large public repositories and inked
together in order to create axioms. The axioms can describe any kind of level of abstraction;
therefore the methodology can scale across various biological levels, from the molecule to
the phenotypic trait. Open Biomedical Ontologies have a Web Ontology Language (OWL)
representation, which enables the use of tools and reasoning engines coming from the
semantic web community. An OWL reasoner running over the axioms would provide a proof
of concept for the new effect of the drug on a particular phenotype, in an automated way.
Research Questions
What is the minimum and necessary set of axioms that enables the representation of the
biological system?
Are there enough quality facts in biomedical databases in order to make useful deductions?
What are the advantages brought by a reasoning engine?
How can we represent the effect of a drug in description logic? Can we prove the correctness
of an indication for an active compound?
How can it be evaluated? Can we suggest new indications for already approved drugs?
Research Methodology
Activity abstraction layer: The medical value of a drug is tightly bound to the subsequent
activity that it has on a biological body. In order to ease the integration of the various
biological layers, we argue for a representation at a common semantic level, scaling up from
chemicals to phenotypes. Considering the activity of the members of a system enables such
a representation. In other words, our methodology does not represent a molecule for its
physical structure, but considers rather the biological function carried by this molecule. We
refer to the activity born by a compound by adding some curly brackets around the
compound name. For instance the activity of a compound A is written {A} (Figure 1). The GO
provides definitions and annotations which we re-use, in order to describe the abstract
activity born by a biological entity.
Knowledge integration: The common level of representation of the actors involved in the
system as activities eases the integration ask. Indeed, abstract molecular activities can be
linked together using the positive/negative regulation relationships provided by the Relation
Ontology (RO), in order to create triples. It is also possible to furthermore link phenotypic
processes as described in the GO using these same relations. We argue that these two
relations are intuitive and accurate enough to capture dependencies mong he molecular
activities of a biological system. The triples, which we refer as axioms or facts, are generated
from the content of pathway databases: We convert the chemical reactions representation
into activity dependencies relationships (Figure 1). The connection of facts creates an activity
map, here the nodes are activities/processes and the edges are relations from the RO.
Figure 1: Conversion of an enzymatic reaction as
represented in pathway databases into activity
dependencies axioms, sing he positive/negative
regulation from the RO.
Reasoning: The activity map is represented in OWL in order to re-use the reasoning
frameworks already developed. The reasoner provides a mean to check the consistencies of
the integrated facts that are building the activity map. It also computes the causal effects of
drugs acting on the map: In biological systems, a drug disturbs first the activity of a molecular
target, and then a perturbation wave follows, influencing the activities and processes linked
to the original target. The reasoner renders this effect by changing the states of activity of the
nodes of the activity map that are directly or indirectly connected to the drug target. The
nodes are either becoming positively or negatively perturbed, according to the state of the
previous nodes to which they are linked to, and to the type of edge chaining the two nodes
(positive or negative regulation). From this series of logical steps, the reasoner can
demonstrate and explain the causal influence an active compound has on a large set of
integrated molecular activities and phenotypic processes.
References
1. Thor, K. B. & Katofiasc, M. A. (1995), Effects of duloxetine, a combined serotonin and norepinephrine
reuptake inhibitor, on central neural control of lower urinary tract function in the chloralose-anesthetized
female cat, in: J. Pharmacol. Exp. Thera. 274.
2. Sanseau P, Koehler J. (2011), Editorial: computational methods for drug repurposing, in: Brief Bioinform.
12(4)