An automated workflow for highly linked and semantically annotated data in catalysis - LARAsuite

Experimental data will only become fully findable and machine actionable, e.g. for ma chine learning and AI applications, if it is properly annotated with metdata following standardised semantics, which are provided by terminologies, taxonomies or ontologies. This enrichment process can become very tedious, time-consuming and requires in-depth knowledge about semantic technologies. This extra burden on scientists reduces the motivation to produce data which is findable, retrievable and machine actionable. One remedy to this is to provide a research data management framework that accompanies a scientist during the research process and adds the required metadata automatically, when the data is produced. The LARAsuite is an implementation of such an open-source research data management framework. In this demonstration, we illustrate an automated workflow to enrich primary experimental data with metadata. This automation is achieved by the powerful API of LARAsuite's (http://gitlab.com/larasuite) database and triple store, which enables fast, script-able interactions and automated generation of a knowledge graph. As a real-life example use-case, we approached a laboratory (BasCat) that was not known during the core LARA development phase to show the generality of the framework. From BasCat we converted gas chromatograms (GC) for syngas to ethanol (StE) hydrogenation reaction from heterogeneous catalysis experiments to highly inter-linked and semantically annotated data. The data was initially acquired in device specific file formats and manually converted and aggregated into spreadsheet tables. These spreadsheets encompass the raw integrated values from GC, as well as processed performance descriptors for the StE reaction. To "simulate" the automated data collection and annotation process, the data of each measurement series was stored in an easy parseable csv file and the metadata, describing the header of the csv file was stored in JSON-LD. In a first step, metadata was enriched by missing information followed by a conversion of the data into the open SciDatS format, which is an apache parquet file with JSON-LD metadata attached directly to the same file (very similar concept as ROCRATE (https://www.researchobject.org/ro-crate/ ). The description of the experiment and the workflow was added as YAML and pythonLab files. These standardised file formats could then be used to feed the rich API of LARA, which in return generated a highly structured and deeply interlinked representation of the complete series of experiments. In the background a knowledge graph is generated in a triple store (openlink virtuoso). The experiments can now be explored via either the LARA web application or via querying the triple store with SPARQL. With a click of a button, the full dataset can be synchronised with an external repository, like dataverse, CKAN or Zenodo. In our case, we tested uploading the data to Repo4Cat (NFDI4Cat data repository).