AOP-DB methodology and OECD Project for Improved AOP Reporting Standards for Biomedical Data

This is a European commission, Joint Research Council invited presentation to discuss current AOP initiatives and activities regarding the mapping and interoperability of AOP associated biomedical information (gene, protein, disease, phenotype, etc) to support improved AOP data standards for future New Approach Methodologies (NAMs), and AI related tasks. The meeting takes place in Ispra, IT. Dr. Mortensen will present virtually.

Impact/Purpose

The concept of annotating KEs is to some extent incorporated into the AOP-Wiki via Key Event Components (KECs, e.g., Gene Ontology (GO) terms) and third-party tools, however, these annotations are too sporadic and generic to systematically represent KEs as the corresponding genes, proteins or metabolites. Recent work from the group of Professor Dario Greco (Saarimäki et al., 2023a, 2023b) has demonstrated the value of annotating sets of genes to KEs by means of token-based similarity (via techniques of Natural Language Processing (NLP)) and manual curation. This approach inspired creation of Omics2AOPs, and its formalisation as an OECD-endorsed project to build and expand on the original approach by working with a group of cross-disciplinary experts in areas including AOPs, toxicology, omics, and Large Language Models (LLM).   Objectives of Omics2AOPs  a. Establishment of methodology for annotating sets of genes, proteins and metabolites to KEs using high-quality knowledgebases and ontologies. These knowledgebases and ontologies are a bridge to map genes/proteins/metabolites onto KEs/AOPs. Omics2AOPs will:  • expand the input data, when compared to the approach of Saarimäki et al. (2023a, 2023b)  to enable annotations of sets of genes, proteins or metabolites to KEs and AOPs; • include AOPs relevant to both human- and eco-toxicology; • allow for greater automation: e.g., by updating/expanding the techniques of NLP for the annotations of molecule sets to KEs, including (if appropriate) LLMs to support the annotation process; • utilise the methodology to perform the full annotation of available KEs. b. Development of a tool that implements the methodology to execute annotation. • The project’s primary output (in addition to the methodology, described above) is an open-source tool that the community may use to perform and access these annotations. Establishing the methodology and development of such tool must occur in parallel for the tool to be fit-for-purpose.  • Format of the tool: with respect to the implementation of the methodology from objective 1, although several options exist, a Python/R Package with GUI may offer the necessary computational capabilities to perform such annotations and a userfriendly interface without the need to develop a standalone software application (at least at this stage of the project). In addition, a website may also be a user-friendly 1 solution for accessing the annotations (e.g., an output of a Python/R package) and/or the tool itself. c. Evaluation and demonstration of the utility and usability of such tool through use cases. • The tool (and therefore, the methodology) will be trialled to evaluate its performance through use cases. One immediate example is performing functional enrichment analysis for transcriptomics, proteomics, and metabolomics (experimental) data. 4. Presentations and topics of discussion planned for the workshop a. Activities relevant to Omics2AOPs with presentations (20 min + 10 min Q&A) from:  • Dario Greco & Alexandra Schaffert (Tampere University) on the original approach of Saarimäki et al. (2023a, 2023b), including the methodology, and further developments by the group;  • Giulia Callegaro (Leiden University) on the EFSA-funded TXG-MAP project and mapping of co-expression network models to AOPs; • Marvin Martens (Maastricht University) on mapping WikiPathways to Key Events, including the methodology; • Holly Mortensen (US EPA, online) on AOP-DB, including the methodology, as well as the project on AOP Reporting for Biomedical Data;  • Mark Viant (University of Birmingham) on mapping knowledgebases to create a panel of metabolic biomarkers for toxicology (MTox700+) and further developments in this area; • Ginnie Hench (Open BioData Modeling, LLC) on AOP Event Component-based text annotation, including the use of NLP; • Harry Caufield (Lawrence Berkeley National Laboratory) on ontology extraction using Large Language Models (including currently available approaches).

Citation

Mortensen, H. AOP-DB methodology and OECD Project for Improved AOP Reporting Standards for Biomedical Data. Omics2AOPs – workshop and kick-off meeting, Ispra, ITALY, October 22 - 24, 2025.