Biomedical Event Extraction for Chemical Risk Assessment

German. European agriculture makes widespread use of chemical pesticides to optimise crop yields, with pesticides sales around 350,000 tonnes per year (from 2011 to 2020). Exposure to pesticides has been linked to severe health issues in humans including cancer, heart, respiratory and neurological diseases. Moreover, they are known to reduce biodiversity in insect species and impact ecosystem health, putting food production at risk in the future1. Chemical risk assessment aims to characterise potential effects of a hazard on biological and ecological systems. Crucial effects, so-called key events (KE) of a substance, that are observed on molecular, cellular and organ level and subsequent negative overall impact on the organism (e.g. cancer) are modelled in Adverse Outcome Pathways (AOPs) in the AOPwiki2. New Approach Methodologies (NAMs), a wide range of animal-free approaches, provide very large amount of data for these events. For a specific risk assessment question experimental evidences, that are stored in various data sources have to be evaluated in an elaborate review process. Therefore we aim to automatically extract key event components consisting of the biological process (e.g. gene expression), a biological object (e.g. a gene), and the direction of the perturbation of the system (increased or decreased) from publications. Event extraction is a popular approach to resolve the complex interactions of biomedical entities in texts. We detect entities and events across multiple levels of biological organisation from molecular to the organ system level in a dataset with focus on toxicity of polychlorinated Dibenzodioxins and Dibenzofurans (PCDD/PCDF) from the AI4NAM project3. This research has been funded by the German Federal Ministry of Food and Agriculture (BMEL) in the research project “KI- & Daten-Akzelerator (KIDA)” with project number 28KIDA004