The Web was designed to improve the way people work together. The Semantic Web extends the Web with a layer of Linked Data that o ers new paths for scienti c publishing and co-operation. Experimental raw data, released as Linked Data, could be discovered automatically, fostering its reuse and validation by scientists in di erent contexts and across the boundaries of disciplines. However, the technological barrier for scientists who want to publish and share their research data as Linked Data remains rather high. We present two real-life use cases in the elds of chemistry and biology and outline a general methodology for transforming research data into Linked Data. A key element of our methodology is the role of a scienti c data curator, who is pro cient in Linked Data technologies and works in close co-operation with the scientist.
The World Wide Web was envisioned by its inventor Tim Berners-Lee as a
universal information space that enables people to work together and collaborate
better [
A cornerstone of the scienti c method is the requirement that any experiment
has to be reproducible [
This requires that the data is su ciently annotated and documented so that the exact experimental conditions can be identi ed.
Because scienti c research data is very valuable, funding agencies have a high
interest to prevent duplication of e ort and foster the reuse existing data as e
ciently as possible [
In general, scientists have been rather reluctant to adopt Semantic Web
technologies. The reasons for this reluctance were revealed by several surveys (for a
summary cf. [
But there is a growing number of scientists who share the ideal of making research data public and are willing to publicly release their data. These early adopters face another barrier in the form of technical complexity. A considerable e ort is necessary to get acquainted with the relevant techniques and paradigms, i.e. Semantic Web and Linked Data technologies. In order to learn more about possible ways to overcome this obstacle, we investigated real-life use cases from natural science departments of our university. We interacted with scientists at our university, and developed a rst methodology targeted at lowering the barrier for scientists to release their research data as Linked Data.
Our long-term goal is to develop and validate a methodology with appropriate tools support that facilitates the task of publishing research data as Linked Data as well as to assess and compare the cost, feasibility and ease of use of di erent approaches systematically. In this paper we describe two use cases we are currently implementing. Using these as a springboard we will explore the promises and possible pitfalls of publishing scienti c research as Linked Open Data. 2
The main objective of researchers is to provide answers to open scienti c questions in their eld, thus advancing their own understanding of key problems and phenomena as well as the one of their research eld as a whole. Taking on the additional workload of semantically annotating research data will only be considered if it does not put too much strain on their time budget and can be integrated with their research work. To overcome this obstacle we decided to investigate an approach of co-operation and support. We contacted scientists who are willing to share their data and o ered to take care of the technical side of the publication of research data as Linked Data while the scientists contribute their domain knowledge.
We selected two current research projects carried out by natural science departments from Bielefeld University: one from chemistry and one from biology. Both topics are highly interdisciplinary and produce research data that is potentially relevant to researchers from other disciplines. Both scientists were open to the idea of sharing their research data and were willing to contribute their domain knowledge. In the following we will present these two use cases as well as the involved scientists in more detail. 2.1
Thomas Koop is a professor of Physical Chemistry at Bielefeld University
(Germany). He is co-founder and executive editor of the open access journal
Atmospheric Chemistry and Physics 1. His research interests include the properties of
atmospheric aerosols and their in uence on cloud formation. In September 2011,
he published a paper on the glass transition of organic aerosols [
Aerosols, which consist of oating particles in the air, are an important
factor in many atmospheric processes, like light scattering and cloud formation.
According to new insights, water soluble organics can form amorphous solids
(glasses) in the upper troposphere (i.e. at 8-15 km height), which inhibit ice
crystal formation, thereby a ecting cirrus cloud formation [
In order to quantify the magnitude of this e ect, data about the glass transition temperature Tg of various substances known to be present in the atmosphere is needed. Because glass transition temperatures are not collected in chemical databases, Thomas Koop conducted an extensive, manual literature research, which took about 100 hours of work. He collected the resulting 596 Tg values from 22 publications in a large spreadsheet-table and supplemented them by additional information like provenance, measurement methods and additional comments.
The corresponding publication [
Use Case: As a use case we take the example of a researcher in chemistry who wants to collect glass transition temperatures of aerosols. Instead of compiling the data manually from published research articles as Thomas Koop did, our scientist would use Semantic Web search engines to collect relevant data and use
appropriate SPARQL queries to aggregate results as needed. Issues that need to be paid attention to are provenance, data quality as well as the fact that di erent vocabularies might have been used in publishing the data, so that vocabulary harmonization is a crucial part of the process.
Some sample competency questions a researcher could pose to the dataset are: Q: Give me all glass transition temperatures of organic compounds! Q: Give me all glass transition temperatures of amino acids measured by differential scanning calorimetry! Q: Which substances form glasses at temperature and
pressure conditions in the troposphere? Volker Durr is a professor of Biological Cybernetics at Bielefeld University (Germany). His research interests include the question of how insects adapt their locomotion behaviour to the context of the situation. He coordinates the EU project EMICAB2, which has the objective to develop an autonomous hexapod robot.
Insects like the stick insect (Figure 2) can walk on rough terrain, climb
obstacles, and use their legs for other behavioural tasks such as searching or reaching
[
resource-e cient information processing for solving complex behavioural tasks,
the analysis and modelling of insect locomotion have been proposed as a basis
for improving arti cial autonomous walking robots [
The movement of stick insects can be measured by marker-based motion capturing: markers are attached to the body of the insect and tracked by an infrared camera system. The resulting trajectories (time-ordered xyz -coordinates) describe the movement of the insect in space. Volker Durr's group recorded several hours of locomotion sequences from di erent stick insect species by motion capture. The interpretation of the trajectory data is dependent on the body morphology and the position of the markers on the body. Motion capture datasets have been released in the past, but without specifying the anatomy of the test subject and the exact marker locations, such that these datasets are of limited use outside their original purpose.
A novel approach is to provide su cient annotation for calculating joint angle time courses for all degrees of freedom from the trajectory data. This would allow the data to be interpreted and reused in other contexts. Pioneering this approach, the EU project EMICAB will make such calculated data publicly available alongside the experimental raw data and metadata about the experimental conditions under which the data was obtained. A semantic annotation of these datasets would greatly improve their retrieval and interpretation by potential future users.
Use Case: A researcher interested in insect motion might download this dataset and extract or recompute the joint angle time courses for all degrees of freedom, thus being able to simulate the organism or compare it to his own results for other or the same organism. The challenge is to incorporate enough information in the data about how the joint angle time courses have been computed so that the comparison is meaningful.
Competency questions the dataset has to answer include: Q: Give me all motion capture datasets about insects! Q: How large is the complete dataset? Q: What data is necessary to reproduce the experiment? Fig. 2: Stick insect movement with markers for motion capture attached. (Reproduced with permission of Volker Durr)
A key element in our methodology is the role of the scienti c data curator. The data curator's task is to translate the methods and results of scienti c research into Linked Data. His role could be compared to that of an investigative reporter: he is not an expert in the domain he is describing, but he is pro cient at nding out what is essential and relevant. He asks the scientist the right questions to nd out what others need to know to understand and reuse the data. Further, he should be pro cient in Semantic Web and Linked Data technologies.
Transferring scienti c research into Linked Data can be viewed as a project which requires a joint e ort between the scientist and the data curator, such that a close co-operation and constant feedback is essential during all phases of the project. We propose a methodology which involves seven consecutive tasks:
The kick-o meeting is the rst meeting of the scientist and the data curator and marks the start of the project. In the kick-o meeting the project members get to know each other and lay the groundwork for the future co-operation. The data curator interviews the scientist about his research interests and gives an introduction into the technology of Linked Data. Ideas and expectations are exchanged in order to build a common understanding of the goal and scope of the project, which will be de ned in the next step.
Following the kick-o meeting, the data curator formulates a proposal for the goal and the scope of the project and subsequently re nes it by feedback from the scientist. As a main tool at this stage of the project we formulate competency questions, which can be used as tests to make sure that the data contains all relevant information, and to choose vocabularies to represent the data.
For our use cases the goal is to capture all relevant data, i.e. the experimental results and all information necessary to reproduce these results. A more lightweight approach could concentrate only on the data essential for interpreting the experimental results. The most comprehensive scope would be to use all available data, even the pieces that seem irrelevant for the reproducibility of the experiment - but could prove relevant in the future or in other contexts.
The data curator acquires domain speci c knowledge. He achieves this by interviews with the scientists and reading the papers which are based on the experiments. His aim is not to become an expert himself but to get an overview and basic understanding in a short period of time. In addition he collects data which might already be available in structured or semi-structured form.
The glass transition temperatures were collected in a large spreadsheet table with informal comments and undocumented color-coding. The stick insect movement was available in a relational database.
At this stage, the data curator explores existing vocabularies and ontologies that could be reused. He has to thoroughly investigate them in order to evaluate their applicability and usefulness for his task. In addition he is looking for existing Linked Open Data (LOD) datasets that can be linked to. Interlinking and reuse is extremely important, because most of the usefulness of the data lies in its connection to external data. If concepts or resources are involved for which no existing vocabularies or datasets can be located, the data curator will create them. The understanding, evaluation, disambiguation and alignment of existing ontologies is the most important and labour intensive task of the whole process because many existing ontologies are not properly documented.
For our use case in chemistry, several ontologies for the domain of chemistry
exist, e.g. CHEMINF [
If one or more ontologies have been selected, the data is encoded using the technological tool most appropriate. This could range from the mapping of an existing database, using annotation software or even manual encoding. Figure 3 presents sample RDF code for encoding a glass transition temperature. :PinicAcid a :ChemicalSubstance , :hasCASNumber "[473-73-4]" ; :hasName "Pinic Acid"@en ; :hasProperty [ a :GlassFormationPoint ; dc:source "http://dx.doi.org/10.1039/C1CP22617G" ; :hasValue "268.1"^^xsd:float ; :hasUnit :Kelvin ; :hasStandardDeviation "4.8"^^xsd:float ; :hasMeasurementCondition [ a :MeasurementPressure ; :hasValue "101325"^^xsd:float ; :hasUnit :Pascal ] ; :hasExperimentalTechnique :differentialScanningCalorimtery ] .
Fig. 3: RDF-representation of the glass transition temperature of pinic acid.
The data is published, either by uploading the code representing the knowledge to a web server or by importing it into a triplestore. This task essentially completes the project. A SPARQL endpoint should be also provided ideally so that the data can be queried exibly as needed.
Subsequently, the usage of the published data is continuously monitored, for example by looking at SPARQL queries generated by third parties. This can help to improve and re ne the data selection and implementation.
Documenting is not a separate task but is done parallel to the other tasks. Like in all projects, documentation plays an important role in forming a common understanding between project members, to proceed from one task to the next, and to enable others to understand and continue the work in the future. The individual tasks are not strictly linear but feedback loops to earlier tasks are possible if a subsequent task should require correction or re nements to an earlier task. Figure 4 shows an overview of the proposed methodology and possible feedback loops between the tasks.
The Open Science movement aims to make scienti c publications and research data publicly available. Numerous initiatives have formed over the last few years that put these ideals into practice. Open access journals create alternatives to the old publication system, e.g. Atmospheric Chemistry and Physics4, which has been using an open review system for 10 years and has the highest impact factor of all 68 journals in the eld of meteorology and atmospheric sciences. Even traditional publishers are beginning to embrace the new technologies, like Elsevier did with its Grand Challenge5. Universities are setting up public repositories of research data, e.g. VIVO6 or Potsdam Mind Research Repository 7, which gives access to peer-reviewed publications and additional data and scripts for analyses and gures. Large Datasets have been opened, like the Human Genome Project 8 or the Sloan Digital Sky Survey9. The W3C's Health Care and Life Sciences Interest Group (HCLSIG)10 created a knowledge base of RDF data from the domains of health care and the life sciences. Social networks like myExperiment11 allow scientists to publish and share their scienti c work ows. With all of these the ideas of Open Science are gradually changing the way scienti c research is done.
One of the main tasks of our methodology is the elicitation of knowledge from
the domain experts. Methodologies for knowledge extraction have a long
tradition in knowledge management (cf. [
For an e cient creation of Linked Data several approaches have been developed, either by automated translation or by tool-support for the author. Four kinds of approaches can be distinguished: 1. Export from existing sets of structured data: relational databases, like the ChEBI database12, which collects data about chemical substances, are exported into Linked Data by mapping database elds to a vocabulary. The D2R Project13 exposes the content of a relational database as Linked Data. 4 http://atmos-chem-phys.net/ 5 http://www.elseviergrandchallenge.com/ 6 http://vivoweb.org/ 7 http://read.psych.uni-potsdam.de/pmr2/ 8 http://www.ornl.gov/sci/techresources/Human_Genome/home.shtml
10 http://www.w3.org/wiki/HCLSIG
11 http://www.myexperiment.org/
12 http://www.ebi.ac.uk/chebi/
13 http://d2rq.org
2. Export from content management systems (CMS): Drupal14, WordPress15
can publish editorial content as Linked Data using pre-selected vocabularies.
3. Automated extraction of data from scienti c publications by text-mining
techniques: Several methods for automatic extraction of bibliographic
metadata have been developed [
Which of these approaches is the best one for a speci c task depends on the goal and the scope of the individual project. Because our goal is to develop deeper insights into how existing vocabularies and ontologies can be reused in the process of publishing Linked Data, we have decided to carefully evaluate and select the most appropriate vocabularies instead of converting the data to RDF using some automatic approach (e.g. RDB2RDF17). In the future we plan to compare the results with those of more automatic approaches. 5
We presented two use cases from chemistry and biology that we are currently working on. The speci c aim of these projects is to publish the relevant scienti c research data as Linked Data, i.e. the results of the experiments and the experimental set-ups necessary to reproduce the results. We proposed a methodology that is characterized by a close co-operation between a scientist and a scienti c data curator, who translates the scientists' domain knowledge into Linked Data. This preliminary methodology will be validated and re ned empirically as the implementation progresses.
In preparing the projects we found that all scientists we interacted with are interested in the ideas and possibilities of Linked Open Data. But only few of them are willing to contribute and invest their data, time and knowledge. A close co-operation between scientist and data curator is highly important for the success of the project. Therefore trust is essential. The scientist must be sure that his data is handled responsibly and that his wishes regarding its publication are respected.
So far we have de ned the goals and the scope of both projects and elicited the relevant domain knowledge. We are currently in the process of evaluating suitable existing ontologies and Linked Data from other datasets we could link to. Because our goal to publish all relevant research data is rather ambitious, the cost involved with each of the steps is high. Especially the exploration and evaluation of existing ontologies has proven to be complex and time consuming.
Our long-term objective is to contribute to the formation of an open research infrastructure by empowering scientists to publish their research as Linked Data. Towards this goal, appropriate methodologies for the transformation of research data into Linked Data are needed. In combination with shared ontologies and tool support, we expect these to be the foundation for scientists to adopt the new technology of Linked Data. Our hypothesis is that the role of a scienti c data curator as proposed in this paper is a key function towards facilitating this development.
After completing the two use cases we will perform a thorough analysis of the resulting datasets and of the overall process. Focus will be put on the question of the cost involved for each of the tasks. Our next step will be the development of criteria for choosing, combining and expanding existing ontologies. In future work we plan to use our manually created Linked Data as a gold standard for the evaluation of less expensive, semi-automatic or fully automatic solutions.
We are grateful to Thomas Koop and Volker Durr for sharing their research data and their helpful insights.
This work is funded as part of the Center of Excellence Cognitive Interaction Technology (CITEC) at Bielefeld University.