Data collation and curation are critical steps in modern data-driven science. Since the emergence of manifold open data sources in the life sciences domain, data availability is no longer a major obstacle in (even academic) drug-discovery projects. Integrated platforms - such as the Open PHACTS discovery platform – are semantically linking diverse (mainly pharmacological) data sources which facilitates data access, its inter-connectivity, and re-use. Exemplars of this kind can further be used as a prototype for building up integrated data platforms in other (life science) domains, where data heterogeneity is still a main challenge and prevents its full exploitation (such as toxicological data). In biomedical research projects, recent challenges include the proper handling of the data for the use to answer the underlying research questions. Nowadays, datasets no longer have to be tediously collected from the medicinal chemistry literature by hand, but more often are semi-automatically extracted from open data sources i.e. by the use of elaborated workflows. Within these workflows, proper filtering and cut-off setting are having a major impact on data quality, validity, and completeness. Recently, we have shown that intra-variabilities (same compound target pair, same bioactivity endpoint, different assays) seem to be in the same range as inter-variabilities (same compound target pair, different bioactivity endpoints, different assays) for IC50 and Ki measurements of the human serotonin and dopamine transport inhibitors. In the same study we also demonstrated that cut off setting for labelling actives/inactives shall better be tailored to the specific protein target under study, and activity endpoint (if multiple different assay setups are mixed). From our perspective, the biomedical community is lacking clear guidelines as to the proper usage of different types of data for subsequent predictive modelling. Depending on the nature, quality and the granularity of the data used, certain modelling techniques (e.g. structure- activity relationships, predictive classification models, modelling of in vitro/in vivo toxic effects) might be permissible or not. In the presentation we will summarize recent challenges in data-driven drug discovery, give examples from our work, and line out ongoing initiatives designed to address those challenges.

Proper data extraction and curation in an era of linked open data – Are we there yet?

253rd National Meeting (2017)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.