Innovations developed within eTRANSAFE have been selected for publication by the European Commission’s Innovation Radar.
The Innovation Radar, a European Commission initiative to identify high potential innovations and innovators in EU-funded research and innovation framework programmes, has selected three innovations of the eTRANSAFE project to be included on the Innovation Radar platform:
- Preclinical text mining solution for treatment response
- The Rosetta Stone – bridging the gap between preclinical and clinical data vocabularies
- Advanced in silico modelling to predict toxicity
These innovations, stemming from the project work, have been classified by the Innovation Radar as Exploring innovations in the early phases of technological readiness and have been published on the EU-funded innovations website on July 20th 2020, thus joining the 3600+ EU-funded innovations already showcased on the platform.
Preclinical text mining solution for treatment response
A Preclinical Text Mining Pipeline has been developed to capture expert conclusions for treatment-related findings from animal toxicology study reports into the eTRANSAFE Preclinical Database in a consistent, structured, machine-readable format. This innovation was led by Hospital del Mar Medical Research Institute (IMIM) and Barcelona Supercomputing Center (BSC) in collaboration with Lhasa Limited and PDS Consultants and includes text mining tools aimed at extracting expert conclusions from animal toxicology study reports into a novel “Study Report (SR)-Domain” template, as a proposed additional domain to SEND, for integration into the cloud-hosted eTRANSAFE Preclinical Database Platform. Read More
The Rosetta Stone – Bridging the gap between preclinical and clinical data vocabularies
The analysis of the predictive value of preclinical safety data for the human safety evaluation requires the alignment and harmonization of terminologies of the two areas. Each area has its own standardization (e.g. SEND, INHAND for preclinical toxicology or ICD, MedDRA, SNOMED CT for clinical data) but a connection between the two worlds, allowing translational big data analyses, has been missing. The Rosetta Stone project is striving at establishing such translation and Erasmus Universitair Medisch Centrum Rotterdam (EMC), in collaboration with the pharma industry, has developed advanced terminology mapping to carry out translational analysis from preclinical to human data, and vice versa. SNOMED CT is used as a bridge between SEND (preclinical) and MedDRA (clinical) terminologies. Through automated and expert curation methods, this ‘Rosetta Stone’ will enable holistic analysis of toxicity for comparing findings from in vivo studies with adverse events observed in humans that will go well beyond the currently limited concordance analyses attempted.
Advanced in silico modelling to predict toxicity
Universitat Pompeu Fabra (UPF) has developed Flame, an advanced modelling framework for the development and application of predictive models. This software allows to develop machine learning or other types of predictive models and encapsulate them in self-contained predictive engines, which can be easily transferred by Internet. This facilitates the collaboration between academic and industrial partners, since models generated using open data can be easily distributed and used in production. Moreover, models installed locally can be easily re-trained with corporate (often confidential) data to further improve their quality.
Flame incorporates all the tools required to build predictive models starting from a collection of annotated chemical structures and supports natively most widely used molecular descriptors and machine learning methods. It also incorporates innovative model combination strategies for supporting multi-level modelling. Flame is a web-application with a user-friendly graphical interface which can run in a desktop computer or be deployed in a server to run predictions in all the computers connected to the company intranet. This modelling framework is being used for building predictive models for different types of toxicological endpoints and can even be applied to predict other biological properties of interest.